Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileWorks460 from public data
- Learning to Compare: Relation Network for Few-Shot Learning4,817
A conceptually simple, flexible, and general framework for few-shot learning, where a classifier must learn to recognise new classes given only few examples from each, which is easily extended to zero- shot learning.
- Fully-Convolutional Siamese Networks for Object Tracking4,457
A basic tracking algorithm is equipped with a novel fully-convolutional Siamese network trained end-to-end on the ILSVRC15 dataset for object detection in video and achieves state-of-the-art performance in multiple benchmarks.
- Rethinking Semantic Segmentation from a Sequence-to-Sequence Perspective with Transformers3,852
This paper deploys a pure transformer to encode an image as a sequence of patches, termed SEgmentation TRansformer (SETR), and shows that SETR achieves new state of the art on ADE20K, Pascal Context, and competitive results on Cityscapes.
- Res2Net: A New Multi-Scale Backbone Architecture3,450
This paper proposes a novel building block for CNNs, namely Res2Net, by constructing hierarchical residual-like connections within one single residual block that represents multi-scale features at a granular level and increases the range of receptive fields for each network layer.
- Conditional Random Fields as Recurrent Neural Networks2,612
A new form of convolutional neural network that combines the strengths of Convolutional Neural Networks (CNNs) and Conditional Random Fields (CRFs)-based probabilistic graphical modelling is introduced, and top results are obtained on the challenging Pascal VOC 2012 segmentation benchmark.
- Point Transformer2,467
The Point Transformer design improves upon prior work across domains and tasks and crosses the 70% mIoU threshold for the first time on the challenging S3DIS dataset.
- 2,466
- Staple: Complementary Learners for Real-Time Tracking1,846
It is shown that a simple tracker combining complementary cues in a ridge regression framework can operate faster than 80 FPS and outperform not only all entries in the popular VOT14 competition, but also recent and far more sophisticated trackers according to multiple benchmarks.
- 1,838
- Fast Online Object Tracking and Segmentation: A Unifying Approach1,528
This method improves the offline training procedure of popular fully-convolutional Siamese approaches for object tracking by augmenting their loss with a binary segmentation task, and operates online, producing class-agnostic object segmentation masks and rotated bounding boxes at 55 frames per second.
- End-to-End Representation Learning for Correlation Filter Based Tracking1,469
This work is the first to overcome this limitation by interpreting the Correlation Filter learner, which has a closed-form solution, as a differentiable layer in a deep neural network, which enables learning deep features that are tightly coupled to the Cor correlation filter.
- Riemannian Walk for Incremental Learning: Understanding Forgetting and Intransigence1,459
This work introduces two metrics to quantify forgetting and intransigence that allow for better insights into the behaviour of IL algorithms and presents RWalk, a generalization of EWC++ and Path Integral with a theoretically grounded KL-divergence based perspective.
- Deeply Supervised Salient Object Detection with Short Connections1,458
This paper proposes a new salient object detection method by introducing short connections to the skip-layer structures within the HED architecture, which takes full advantage of multi-level and multi-scale features extracted from FCNs, providing more advanced representations at each layer, a property that is critically needed to perform segment detection.
- DESIRE: Distant Future Prediction in Dynamic Scenes with Interacting Agents1,131
The proposed Deep Stochastic IOC RNN Encoder-decoder framework, DESIRE, for the task of future predictions of multiple interacting agents in dynamic scenes significantly improves the prediction accuracy compared to other baseline methods.
- 1,085
- 1,021
- Hypergraph convolution and hypergraph attention952
This work introduces two end-to-end trainable operators to the family of graph neural networks, i.e., hypergraph convolution and hypergraph attention, which define the basic formulation of performing convolution on a hypergraph and further enhances the capacity of representation learning by leveraging an attention module.
- GA-Net: Guided Aggregation Net for End-To-End Stereo Matching838
Two novel neural net layers, aimed at capturing local and the whole-image cost dependencies respectively are proposed, which can be used to replace the widely used 3D convolutional layer which is computationally costly and memory-consuming as it has cubic computational/memory complexity.
- The Sixth Visual Object Tracking VOT2018 Challenge Results804
The Visual Object Tracking challenge VOT2018 is the sixth annual tracker benchmarking activity organized by the VOT initiative; results of over eighty trackers are presented; many are state-of-the-art trackers published at major computer vision conferences or in journals in the recent years.
- 790
- The Visual Object Tracking VOT2016 Challenge Results776
The Visual Object Tracking challenge VOT2016 goes beyond its predecessors by introducing a new semi-automatic ground truth bounding box annotation methodology and extending the evaluation system with the no-reset experiment.
- The Development and Comparison of Robust Methods for Estimating the Fundamental Matrix775
A variety of robust methods for the computation of the Fundamental Matrix, the calibration-free representation of camera motion, are developed from the principal categories of robust estimators, viz. case deletion diagnostics, M-estimators and random sampling, and the theory required to apply them to non-linear orthogonal regression problems is developed.
- GDumb: A Simple Approach that Questions Our Progress in Continual Learning753
A general formulation for the Continual Learning problem for classification—a learning task where a stream provides samples to a learner and the goal of the learner, depending on the samples it receives, is to continually upgrade its knowledge about the old classes and learn new ones is discussed.
- 706
- Siam R-CNN: Visual Tracking by Re-Detection678
This work presents Siam R-CNN, a Siamese re-detection architecture which unleashes the full power of two-stage object detection approaches for visual object tracking, and combines this with a novel tracklet-based dynamic programming algorithm to model the full history of both the object to be tracked and potential distractor objects.
- Interactive Image Segmentation Using an Adaptive GMMRF Model644
Estimation is performed by solving a graph cut problem for which very efficient algorithms have recently been developed, however the model depends on parameters which must be set by hand and the aim of this work is for those constants to be learned from image data.
- The Visual Object Tracking VOT2014 Challenge Results606
The Visual Object Tracking challenge 2014, VOT2014, aims at comparing short-term single-object visual trackers that do not apply pre-learned models of object appearance and introduces a new dataset with full annotation of targets by rotated bounding boxes and per-frame attribute.
- 553
- Continual learning with tiny episodic memories504
It is observed that a very simple baseline, which jointly trains on both examples from the current task as well as examples stored in the memory, outperforms state-of-the-art CL approaches with and without episodic memory.
- Learning feed-forward one-shot learners500
This paper constructs the learner as a second deep network, called a learnet, which predicts the parameters of a pupil network from a single exemplar, and obtains an efficient feed-forward one-shot learner, trained end-to-end by minimizing a one- shot classification objective in a learning to learn formulation.
- 496
- Managing extreme AI risks amid rapid progress478
Drawing on lessons learned from other safety-critical technologies, this work outlines a comprehensive plan that combines technical research and development with proactive, adaptive governance mechanisms for a more commensurate preparation of AI safety research.
- Global Texture Enhancement for Fake Face Detection in the Wild477
A new architecture coined as Gram-Net, which leverages global image texture representations for robust fake image detection and generalizes significantly better in detecting fake faces from GAN models not seen in the training phase and can perform decently in detectingfake natural images.
- 3D Hand Shape and Pose From Images in the Wild424
This work presents the first end-to-end deep learning based method that predicts both 3D hand shape and pose from RGB images in the wild, consisting of the concatenation of a deep convolutional encoder, and a fixed model-based decoder.
- 417
- Learning disentangled representations with semi-supervised deep generative models380
This work proposes to learn disentangled representations that encode distinct aspects of the data into separate variables using model architectures that generalise from standard VAEs, employing a general graphical model structure in the encoder and decoder.
- On the Robustness of Semantic Segmentation Models to Adversarial Attacks352
This paper presents what to their knowledge is the first rigorous evaluation of adversarial attacks on modern semantic segmentation models, using two large-scale datasets and shows how mean-field inference in deep structured models and multiscale processing naturally implement recently proposed adversarial defenses.
- The Eighth Visual Object Tracking VOT2020 Challenge Results350
A significant novelty is introduction of a new VOT short-term tracking evaluation methodology, and introduction of segmentation ground truth in the VOT-ST2020 challenge – bounding boxes will no longer be used in theVDT challenges.
- PhysFormer: Facial Video-based Physiological Measurement with Temporal Difference Transformer343
This paper proposes the PhysFormer, an end-to-end video transformer based architecture, to adaptively aggregate both local and global spatio-temporal features for rPPG representation enhancement, and proposes the label distribution learning and a curriculum learning inspired dynamic constraint in frequency domain.
- Multi-agent Diverse Generative Adversarial Networks335
MAD-GAN, an intuitive generalization to the Generative Adversarial Networks and its conditional variants to address the well known problem of mode collapse is proposed and its efficacy on the unsupervised feature representation task is shown.
- Recurrent Instance Segmentation335
This work proposes a new instance segmentation paradigm consisting in an end-to-end method that learns how to segment instances sequentially, based on a recurrent neural network that sequentially finds objects and their segmentations one at a time.
- 327
- ManiGAN: Text-Guided Image Manipulation324
A novel generative adversarial network (ManiGAN), which contains two key components: text-image affine combination module (ACM) and detail correction module (DCM), which selects image regions relevant to the given text and then correlates the regions with corresponding semantic words for effective manipulation.
- 312
- Online Real-Time Multiple Spatiotemporal Action Localisation and Prediction308
This work presents a deep-learning framework for real-time multiple spatio-temporal (S/T) action localisation and classification that is not only capable of performing S/T detection in real time, but can also perform early action prediction in an online fashion.
- MOSE: A New Dataset for Video Object Segmentation in Complex Scenes307
The experiments show that current VOS algorithms cannot well perceive objects in complex scenes, and there are unresolved challenges under complex scenes and more efforts are desired to explore these challenges in the future.
- Language Model Tokenizers Introduce Unfairness Between Languages304
It is shown how disparity in the treatment of different languages arises at the tokenization stage, well before a model is even invoked, and the case is made that future language models should be trained using multilingually fair subword tokenizers.
- Using Hindsight to Anchor Past Knowledge in Continual Learning301
This work complements experience replay with a new objective that is called ``anchoring'', where the learner uses bilevel optimization to update its knowledge on the current task, while keeping intact the predictions on some anchor points of past tasks.
- Hand detection using multiple proposals295
The hand detector exceeds the state of the art on two public datasets, including the PASCAL VOC 2010 human layout challenge and is introduced with a fully annotated hand dataset for training and testing.
- AttentionGAN: Unpaired Image-to-Image Translation Using Attention-Guided Generative Adversarial Networks292
A new Attention-Guided Generative Adversarial Networks (AttentionGAN) is proposed for the unpaired image-to-image translation task which can identify the most discriminative semantic objects and minimize changes of unwanted parts for semantic manipulation problems without using extra data and models.
- Deep Deterministic Uncertainty: A New Simple Baseline286
It is found that a single softmax neural net with such a regularized feature- space, achieved via residual connections and spectral normalization, outperforms DUQ and SNGP's epistemic uncertainty predictions using simple Gaussian Discriminant Analysis post-training as a separate feature-space density estimator-without fine-tuning on OoD data, feature ensembling, or input pre-procressing.
- Outlier detection and motion segmentation278
A new method for solving the problem of motion segmentation, identifying the objects within an image moving independently of the background by utilizing the fact that two views of a static 3D point set are linked by a 3 X 3 Fundamental Matrix (F).
- NAPSAC: High Noise, High Dimensional Robust Estimation - it's in the Bag255
The NAPSAC (N Adjacent Points SAmple Consensus) algorithm is derived and its performance is shown to be superior to RANSAC in both high noise and high dimensional spaces.
- Pixelwise Instance Segmentation with a Dynamically Instantiated Network247
An Instance Segmentation system that produces a segmentation map where each pixel is assigned an object class and instance identity label, and far more precise segmentations are achieved, as shown by substantial improvements at high APr thresholds.
- 247
- Higher Order Conditional Random Fields in Deep Neural Networks233
Two types of higher order potentials, based on object detections and superpixels, can be included in a CRF embedded within a deep network to allow inference with the differentiable mean field algorithm and are designed to achieve state-of-the-art segmentation performance on the PASCAL VOC benchmark.
- Graph Inductive Biases in Transformers without Message Passing230
The Graph Inductive bias Transformer (GRIT) is proposed -- a new Graph Transformer that incorporates graph inductive biases without using message passing and achieves state-of-the-art empirical performance across a variety of graph datasets, thus showing the power that Graph Transformers without message-passing can deliver.
- Domain-Invariant Stereo Matching Networks230
This paper proposes a novel "domain normalization" approach that regularizes the distribution of learned representations to allow them to be invariant to domain differences, and a trainable non-local graph-based filter for extracting robust structural and geometric representations that can further enhance domain-invariant generalizations.
- Occluded Video Instance Segmentation: A Benchmark229
A simple plug-and-play module that performs temporal feature calibration to complement missing object cues caused by occlusion is presented, and a remarkable AP improvement on the OVIS dataset is obtained.
- Few-Shot Action Recognition with Permutation-Invariant Attention218
This work builds on a C3D encoder for spatio-temporal video blocks to capture short-range action patterns and outperforms the state of the art on the HMDB51, UCF101, miniMIT datasets.
- Deep Learning for Detecting Multiple Space-Time Action Tubes in Videos217
A huge leap forward in action detection performance is achieved and 20% and 11% gain in mAP are reported on UCF-101 and J-HMDB-21 datasets respectively when compared to the state-of-the-art.
- 214
- Learning to Navigate the Energy Landscape200
This paper presents a novel, general, and efficient architecture for addressing computer vision problems that are approached from an 'Analysis by Synthesis' standpoint, and defines a navigational structure over the solution space that can be used for extremely efficient gradient-free local search.
- XingGAN for Person Image Generation197
A novel Generative Adversarial Network (XingGAN or CrossingGAN) for person image generation tasks, i.e., translating the pose of a given person to a desired one to mutually improve each other is proposed.
- Dual Graph Convolutional Network for Semantic Segmentation.196
The Dual Graph Convolutional Network (DGCNet) models the global context of the input feature by modelling two orthogonal graphs in a single framework, which achieves state-of-the-art results on both Cityscapes and Pascal Context datasets.
- 184
- Long-Term Tracking in the Wild: A Benchmark184
The OxUvA dataset and benchmark for evaluating single-object tracking algorithms is introduced, offering the community a large and diverse benchmark to enable the design and evaluation of tracking methods ready to be used “in the wild”.
- 183
- Weakly- and Semi-supervised Panoptic Segmentation180
A weakly supervised model that jointly performs both semantic- and instance-segmentation and is able to segment both “thing” and “stuff” classes, and thus explain all the pixels in the image.
- Local Class-Specific and Global Image-Level Generative Adversarial Networks for Semantic-Guided Scene Generation167
This work considers learning the scene generation in a local context, and correspondingly design a local class-specific generative network with semantic maps as a guidance, which separately constructs and learns sub-generators concentrating on the generation of different classes and is able to provide more scene details.
- Simultaneous Segmentation and Pose Estimation of Humans Using Dynamic Graph Cuts164
A novel algorithm for performing integrated segmentation and 3D pose estimation of a human body from multiple views by optimizing a cost function based on a Conditional Random Field (CRF).
- 163
- Joint Optimization for Object Class Segmentation and Dense Stereo Reconstruction163
This work provides a flexible framework configured via cross-validation that unifies the two problems of dense stereo reconstruction and object class segmentation and demonstrates that, by resolving ambiguities, which would be present in real world data if theTwo problems were considered separately, joint optimization of the two problem substantially improves performance.
- PhysFormer++: Facial Video-Based Physiological Measurement with SlowFast Temporal Difference Transformer160
This paper proposes two end-to-end video transformer based architectures, namely PhysFormer and PhysFormer++, to adaptively aggregate both local and global spatio-temporal features for rPPG representation enhancement, and proposes the label distribution learning and a curriculum learning inspired dynamic constraint in frequency domain.
- Robust Detection of Degenerate Configurations while Estimating the Fundamental Matrix160
It is demonstrated that proper modeling of degeneracy in the presence of outliers enables the detection of mismatches which would otherwise be missed and is a generalization of the robust estimator RANSAC.
- Dynamic Graph Message Passing Networks157
A dynamic graph message passing network, that significantly reduces the computational complexity compared to related works modelling a fully-connected graph by adaptively sampling nodes in the graph, conditioned on the input, for message passing.
- 155
- Discovering Class-Specific Pixels for Weakly-Supervised Semantic Segmentation154
It is shown that properly combining saliency and attention maps allows for reliable cues capable of significantly boosting the performance, and a simple yet powerful hierarchical approach to discover the class-agnostic salient regions, obtained using a salient object detector, is proposed.
- Large-Scale Unsupervised Semantic Segmentation153
This work proposes a new problem of large-scale unsupervised semantic segmentation (LUSS) with a newly created benchmark dataset to help the research progress and presents a simple yet effective method that works surprisingly well for LUSS.
- Interactive Sketch & Fill: Multiclass Sketch-to-Image Translation152
An interactive GAN-based sketch-to-image translation method that helps novice users easily create images of simple objects and introduces a gating-based approach for class conditioning, which allows for distinct classes without feature mixing, from a single generator network.
- Open Vocabulary Semantic Segmentation with Patch Aligned Contrastive Learning149
Patch Aligned Contrastive Learning (PACL), a modified compatibility function for CLIP's contrastive loss, is introduced, intending to train an alignment between the patch tokens of the vision encoder and the CLS token of the text encoder, to transfer seamlessly to the task of open vocabulary semantic segmentation.
- 149
- Secret Collusion among AI Agents: Multi-Agent Deception via Steganography143
This paper comprehensively formalises the problem of secret collusion in systems of generative AI agents by drawing on relevant concepts from both AI and security literature, and provides extensive empirical results across a range of contemporary LLMs.
- Holistically-Attracted Wireframe Parsing143
This paper presents a fast and parsimonious parsing method to accurately and robustly detect a vectorized wireframe in an input image with a single forward pass, and is thus called Holistically-Attracted Wireframe Parser (HAWP).
- 142
- 142
- Multivariate Relevance Vector Machines for Tracking139
A one-to-many mapping from image features to state space is learned using a set of relevance vector machines, extended to handle multivariate outputs, which reduces the estimation error in clutter compared to shape-context histograms.
- 137
- On-the-Fly Adaptation of Regression Forests for Online Camera Relocalisation136
This paper shows how to circumvent this limitation by adapting a pre-trained forest to a new scene on the fly, and achieves relocalisation performance that is on par with that of offline forests, and the approach runs in under 150ms, making it desirable for real-time systems that require online Relocalisation.
- Anchor Diffusion for Unsupervised Video Object Segmentation128
Inspired by the non-local operators, a technique to establish dense correspondences between pixel embeddings of a reference "anchor" frame and the current one is introduced, which allows the learning of pairwise dependencies at arbitrarily long distances without conditioning on intermediate frames.
- 128
- The Ninth Visual Object Tracking VOT2021 Challenge Results127
The Visual Object Tracking challenge VOT2021 is the ninth annual tracker benchmarking activity organized by the VOT initiative; results of 71 trackers are presented; many are state-of-the-art trackers published at major computer vision conferences or in journals in recent years.
- 125
- Vision Transformer with Progressive Sampling122
An iterative and progressive sampling strategy to locate discriminative regions and when combined with the Vision Transformer, the obtained PS-ViT network can adaptively learn where to look.
- 121
- 119
- 116
- 115
- Efficient Dense Stereo with Occlusions for New View-Synthesis by Four-State Dynamic Programming109
A new algorithm is proposed for efficient stereo and novel view synthesis that synthesises images from a virtual camera in arbitrary position near the physical cameras based on an improved, dynamic-programming, stereo algorithm for efficient novel view generation.
- Fine-tuning can cripple your foundation model; preserving features may be the solution107
A new fine-tuning method called $\textit{LDIFS}$ (short for $\ell_2$ distance in feature space) that, while learning new concepts related to the downstream task, allows a model to preserve its pre-trained knowledge as well.
- 105
- SemanticPaint: Interactive 3D Labeling and Learning at your Fingertips104
SemanticPaint allows users to simultaneously scan their environment whilst interactively segmenting the scene simply by reaching out and touching any desired object or surface, and continuously learns from these segmentations, and labels new unseen parts of the environment.
- SemanticPaint104
SemanticPaint allows users to simultaneously scan their environment whilst interactively segmenting the scene simply by reaching out and touching any desired object or surface, and continuously learns from these segmentations, and labels new unseen parts of the environment.
- Learning to Adapt for Stereo103
This work introduces a ``learning-to-adapt'' framework that enables deep stereo methods to continuously adapt to new target domains in an unsupervised manner and incorporates the adaptation procedure into the learning objective to obtain a base set of parameters that are better suited for un supervised online adaptation.
- Multi-shot Temporal Event Localization: a Benchmark102
This paper proposes a new and challenging task called multi-shot temporal event localization, and accordingly, collects a large-scale dataset called MUlti-Shot EventS (MUSES), which has 31,477 event instances for a total of 716 video hours.
- Real-Time RGB-D Camera Pose Estimation in Novel Scenes Using a Relocalisation Cascade98
An extension of this work that achieves significantly better relocalisation performance whilst running fully in real time, and presents a novel way of visualising the internal behaviour of the forests, and uses the insights gleaned from this to show how to entirely circumvent the need to pre-train a forest on a generic scene.
- 98
- Computationally Budgeted Continual Learning: What Does Matter?95
It is found that, under compute-constrained setting, traditional CL approaches, with no exception, fail to outperform a simple minimal baseline that samples uniformly from memory, suggesting that most existing CL methods are particularly too computationally expensive for realistic budgeted deployment.
- Inducing High Energy-Latency of Large Vision-Language Models with Verbose Images94
It is found that high energy-latency cost during inference of VLMs can be manipulated by maximizing the length of generated sequences by proposing verbose images, with the goal of crafting an imperceptible perturbation to induce VLMs to generate long sentences during inference.
- GeoNet++: Iterative Geometric Neural Network with Edge-Aware Refinement for Joint Depth and Surface Normal Estimation93
A new 3D geometric metric (3DGM) for evaluating depth prediction in 3D, which measures whether the predicted depth can reconstruct high quality 3D surface normals and is a more natural metric for many 3D application domains.
- STEER : Simple Temporal Regularization For Neural ODE91
The proposed regularization technique is simple to implement, has negligible overhead and is effective across a wide variety of tasks and can significantly decrease training time and even improve performance over baseline models.
- DenseCut: Densely Connected CRFs for Realtime GrabCut91
A very fast segmentation technique which still achieves very high quality results is demonstrated and is proposed to replace the time consuming iterative refinement of global colour models in traditional GrabCut formulation by a densely connected crf.
- Mimicking the Oracle: An Initial Phase Decorrelation Approach for Class Incremental Learning89
Class-wise Decorrelation (CwD) is proposed that regularizes representations of each class to scatter more uniformly, thus mimicking the model jointly trained with all classes (i.e., the oracle model), and is simple to implement and easy to plug into existing methods.
- Many Heads Are Better Than One: Improved Scientific Idea Generation by A LLM-Based Multi-Agent System88
This work proposes an LLM-based multi-agent system, i.e., Virtual Scientists (VirSci), designed to mimic the teamwork inherent in scientific research, and investigates the collaboration mechanisms that contribute to its tendency to produce ideas with higher novelty.
- 87
- 85
- 85
- 84
- MALT: Improving Reasoning with Multi-Agent LLM Training77
MALT (Multi-Agent LLM Training), a novel post-training strategy that divides the reasoning process into generation, verification, and refinement steps using a sequential pipeline of heterogeneous agents, and allows each agent to specialize by learning from correct and incorrect trajectories, ultimately improving the end-to-end reasoning chain
- Real-Time Evaluation in Online Continual Learning: A New Hope77
It is shown that a simple baseline outperforms state-of-the-art CL methods under this evaluation, questioning the applicability of existing methods in realistic settings and suggesting that the majority of existing CL literature is tailored to a specific class of streams that is not practical.
- Rethinking Class Relations: Absolute-relative Supervised and Unsupervised Few-shot Learning77
This work rethink the relations between class concepts, and proposes a novel Absolute-relative Learning paradigm to fully take advantage of label information to refine the image an relation representations in both supervised and unsupervised scenarios.
- 77
- 77
- Collaborative Large-Scale Dense 3D Reconstruction with Online Inter-Agent Pose Optimisation75
Using the first system to allow multiple users to collaborate interactively to reconstruct dense, voxel-based models of whole buildings using only consumer-grade hardware, a task that has traditionally been both time-consuming and dependent on the availability of specialised hardware.
- MOSEv2: A More Challenging Dataset for Video Object Segmentation in Complex Scenes74
This work presents MOSEv2, a significantly more challenging dataset designed to further advance VOS methods under real-world conditions, and benchmark 20 representative VOS methods under 5 different settings, highlighting that current VOS methods still fall short under real-world complexities.
- Behind Every Domain There is a Shift: Adapting Distortion-Aware Vision Transformers for Panoramic Semantic Segmentation73
This paper proposes the upgraded Transformer for Panoramic Semantic Segmentation, and enhances theMutual Prototypical Adaptation (MPA) strategy via pseudo-label rectification for unsupervised domain adaptive panoramic segmentation.
- SynthCLIP: Are We Ready for a Fully Synthetic CLIP Training?73
This work provides an analysis on CLIP models trained on synthetic data, providing insights on the data generation strategy, number of samples required, scaling trends, and resulting properties.
- ImageSpirit72
This article proposes treating nouns as object labels and adjectives as visual attribute labels to formulate the image parsing problem as one of jointly estimating per-pixel object and attribute labels from a set of training images, and proposes an efficient (interactive time) solution.
- Unifying Training and Inference for Panoptic Segmentation70
An end-to-end network to bridge the gap between training and inference pipeline for panoptic segmentation, a task that seeks to partition an image into semantic regions for "stuff" and object instances for "things", which is of interest for applications with computation budgets.
- The Thermal Infrared Visual Object Tracking VOT-TIR2016 Challenge Results68
A significant general improvement of results has been observed, which partly compensate for the more difficult sequences introduced into the dataset in VOT-TIR2016.
- The Semantic Paintbrush: Interactive 3D Mapping and Recognition in Large Outdoor Spaces65
An augmented reality system for large scale 3D reconstruction and recognition in outdoor scenes using a purely passive stereo setup, allowing for outdoor use and extended sensing range, and allows the user to draw with a laser pointer directly onto the reconstruction to segment the model into objects.
- Ambiguity Modeling in Latent Spaces65
This paper presents a latent variable model capable of consolidating multiple complementary representations, and extends canonical correlation analysis by introducing additional latent spaces that are specific to the different representations, thereby explaining the full variance of the observations.
- 65
- Bipartite Graph Reasoning GANs for Person Image Generation62
A novel Bipartite Graph Reasoning GAN (BiGraphGAN) block aims to reason the crossing long-range relations between the source pose and the target pose in a bipartite graph, which mitigates some challenges caused by pose deformation.
- An Image Is Worth 1000 Lies: Adversarial Transferability across Prompts on Vision-Language Models61
This work proposes the Cross-Prompt Attack (CroPA), a method that updates the visual adversarial perturbation with learnable prompts, which are designed to counteract the misleading effects of the adversarial image.
- Bottom-up Instance Segmentation using Deep Higher-Order CRFs61
This work focuses on the task of Instance Segmentation which recognises and localises objects down to a pixel level and uses a deep neural network trained for semantic segmentation to reason about instances.
- Large-Scale Binary Quadratic Optimization Using Semidefinite Relaxation and Applications60
This work presents a semidefinite programming (SDP) formulation for BQPs, with two desirable properties, and proposes two solvers, namely, quasi-Newton and smoothing Newton methods, for the simplified dual problem.
- When Do Prompting and Prefix-Tuning Work? A Theory of Capabilities and Limitations59
It is shown that despite the continuous embedding space being more expressive than the discrete token space, soft-prompting and prefix-tuning are potentially less expressive than full fine-tuning, even with the same number of learnable parameters.
- Automatic 3D Modelling of Architecture59
A system which automatically derives 3D models of architectural scenes from multiple images in that it explicitly makes use of strong geometric constraints such as perpendicularity and verticality which are likely to be found in architecture.
- 58
- 58
- Learning Discriminative Space–Time Action Parts from Weakly Labelled Videos58
By using local space–time action parts in a weakly supervised setting, the proposed local deformable spatial bag-of-features in which local discriminative regions are split into a fixed grid of parts that are allowed to deform in both space and time at test-time are demonstrated.
- 57
- BNV-Fusion: Dense 3D Reconstruction using Bi-level Neural Volume Fusion56
This work proposes a novel bi-level fusion strategy that considers both efficiency and reconstruction quality by design, and evaluates the proposed method on multiple datasets quantitatively and qualitatively, demonstrating a significant improvement over existing methods.
- A Scalable Communication Protocol for Networks of Large Language Models55
This work introduces Agora, a meta protocol that leverages existing communication standards to make LLM-powered agents solve complex problems efficiently and observes the emergence of self-organising, fully automated protocols that achieve complex goals without human intervention.
- Straight to Shapes: Real-Time Detection of Encoded Shapes54
This work proposes to regress directly to objects shapes in addition to their bounding boxes and categories, and yields the first real-time shape prediction network, running at ~35 FPS on a high-end desktop.
- Cloth Interactive Transformer for Virtual Try-On53
This article proposes a novel two-stage cloth interactive transformer (CIT) method, aiming at precisely capturing the long-range correlations between the cloth-agnostic person information and the in-shop cloth information, and makes the warped in-shop clothing items look more natural in appearance.
- 53
- The Stochastic Search Network51
A fundamental difficulty when using neural networks applied to pattern recognition, is the problem of stimulus equivalence — the invariance of symbolic information independent of transformation within a search space.
- 50
- 49
- 48
- Let's Take This Online: Adapting Scene Coordinate Regression Network Predictions for Online RGB-D Camera Relocalisation48
The approach replaces the appearance clustering performed by the branching structure of a regression forest with a two-step process that first uses the network to predict points in the original scene, and then uses these predicted points to look up clusters of points from the new scene.
- Learning discriminative space-time actions from weakly labelled videos48
It is demonstrated quantitatively that even with single fixedsized subvolumes, the classification performance of the proposed algorithm is superior to the state-of-the-art BoF baseline on the majority of performance measures, and shows promise for space-time action localisation on the most challenging video datasets.
- 47
- 47
- Cross-Modal Deep Face Normals With Deactivable Skip Connections46
This work proposes a method that can leverage all available image and normal data, whether paired or not, thanks to a novel cross-modal learning architecture that allows learning of a rich latent space that can accurately capture the normal information.
- Fully-trainable deep matching46
This paper rewriting the complete DM algorithm as a convolutional neural network results in a novel deep architecture for image matching that involves a number of new layer types and that, similar to recent networks for image segmentation, has a U-topology.
- 45
- 44
- Sequential Optimization for Efficient High-Quality Object Proposal Generation44
A novel object proposal algorithm, BING++, is proposed, which inherits the virtue of good computational efficiency of BING but significantly improves its proposal localization quality.
- Prototypical Priors: From Improving Classification to Zero-Shot Learning44
This work proposes a way to incorporate visual prototypical concepts as side information in a deep learning framework using prototypes as prior information, and shows how prototypical embeddings incorporated in a conventional convolutional neural network improve the recognition performance.
- Combating Adversaries with Anti-adversaries43
The anti-adversary layer is proposed, which generates an input perturbation in the opposite direction of the adversarial one and feeds the classifier a perturbed version of the input.
- Human Instance Segmentation from Video using Detector-based Conditional Random Fields43
This work proposes a method for instance based human segmentation in images and videos, extending the recent detector-based conditional random field model of Ladicky et.al.
- On New View Synthesis Using Multiview Stereo43
It is shown that application of modern multiview stereo techniques to the newview synthesis (NVS) problem introduces a number of non-trivial complexities, and a synthesis of the two approaches which yields good results on difficult image sequences is addressed.
- 42
- RegMixup: Mixup as a Regularizer Can Surprisingly Improve Accuracy and Out Distribution Robustness42
The effectiveness of the well celebrated Mixup can be further improved if instead of using it as the sole learning objective, it is utilized as an additional regularizer to the standard cross-entropy loss, which significantly improves the quality of the predictive uncertainty estimation of Mixup in most cases.
- 42
- FLIPDIAL: A Generative Model for Two-Way Visual Dialogue42
This work presents FLIPDIAL, a generative model for Visual Dialogue that simultaneously plays the role of both participants in a visually-grounded dialogue, and is the first to extend this paradigm to full two-way visual dialogue (2VD), where the model is capable of generating both questions and answers in sequence based on a visual input.
- 42
- Performance characterization of fundamental matrix estimation under image degradation42
An algorithm for simultaneously estimating the fundamental matrix and corresponding points automatically from the two images is described and the performance of this algorithm is assessed as the images are degraded by JPEG lossy compression.
- Online Continual Learning Without the Storage Constraint41
A simple algorithm, which updates a kNN classifier continually along with a fixed, pretrained feature extractor, selected due to its exceptional suitability for online continual learning, which is able to adapt to rapidly changing streams and has zero stability gap.
- 40
- 40
- 39
- Bottom-Up Top-Down Cues for Weakly-Supervised Semantic Segmentation39
This work considers the task of learning a classifier for semantic segmentation using weak supervision in the form of image labels specifying objects present in the image, and adopts an Expectation-Maximization (EM) based approach, which outperforms the state-of-the-art method with a very high margin.
- 38
- Real-Time Dense Stereo Matching With ELAS on FPGA-Accelerated Embedded Devices38
This work implements and evaluates a field programmable gate arrays (FPGA) accelerated adaptation of the efficient large-scale stereo (ELAS) algorithm, and takes advantage of both components on the CPU/FPGA system-on-chip to showcase the strategy necessary to accelerate more complex and computationally diverse algorithms for such low power, real-time systems.
- A Bayesian Estimation of Building Shape Using MCMC37
This paper investigates the use of an implicit prior in Bayesian Model-based 3D reconstruction of architecture from image sequences and defines an implicit representation that favours global regularities prevalent in architeture.
- 34
- R3SGM: Real-Time Raster-Respecting Semi-Global Matching for Power-Constrained Systems34
This paper proposes a novel, resource-efficient method that is inspired by MGM's techniques for improving depth quality, but which can be implemented to run in real time on a low-power FPGA, and can achieve a state-of-the-art balance between accuracy, power efficiency and speed.
- 33
- 33
- Bayesian Stochastic Mesh Optimization for 3D reconstruction33
A Bayesian formulation in which simplicity and smoothness assumptions are encoded in the prior distribution and the resulting posterior is optimized by simulated annealing, which makes the approach particularly robust to noise and ambiguity.
- 32
- 32
- DeepInteraction++: Multi-Modality Interaction for Autonomous Driving31
This work designs DeepInteraction++, a multi-modal interaction framework characterized by a multi-modal representational interaction encoder and a multi-modal predictive interaction decoder that demonstrates the effectiveness of the proposed modality interaction strategy.
- 31
- Reliability in Semantic Segmentation: Are we on the Right Track?30
This is the first study on modern segmentation models focused on both robustness and uncertainty estimation and it is found that while recent models are significantly more robust, they are not overall more reliable in terms of uncertainty estimation.
- Structure-Preserving 3D Garment Modeling with Neural Sewing Machines30
A novel Neural Sewing Machine (NSM) is proposed, a learning-based framework for structure-preserving 3D garment modeling, which is capable of learning representations for garments with diverse shapes and topologies and is successfully applied to3D garment reconstruction and controllable manipulation.
- Edge Guided GANs with Semantic Preserving for Semantic Image Synthesis30
A novel Edge guided Generative Adversarial Network (EdgeGAN) for photo-realistic image synthesis from semantic layouts that can generate significantly better results than state-of-the-art methods.
- Efficient Continuous Relaxations for Dense CRF30
The advantage of continuous relaxations over the widely used mean-field algorithm on publicly available datasets is demonstrated and a novel divide-and-conquer method to compute the subgradients of a linear programming relaxation that provides the best theoretical bounds for energy minimisation is developed.
- CoMAS: Co-Evolving Multi-Agent Systems via Interaction Rewards29
This work introduces Co-Evolving Multi-Agent Systems (CoMAS), a novel framework that enables agents to improve autonomously by learning from inter-agent interactions without external supervision, thereby enabling decentralized and scalable co-evolution in LLM-based agents.
- Prompting a Pretrained Transformer Can Be a Universal Approximator29
It is demonstrated that much smaller pretrained models than previously thought can be universal approximators when prefixed and offer Jackson-type bounds on the length of the prefix needed to approximate a function to a desired precision.
- Vision Transformers: From Semantic Segmentation to Dense Prediction29
This work explores the global context learning potentials of ViTs for dense visual prediction and forms a family of Hierarchical Local-Global (HLG) Transformers, characterized by local attention within windows and global-attention across windows in a pyramidal architecture.
- kNN-CLIP: Retrieval Enables Training-Free Segmentation on Continually Expanding Large Vocabularies29
This work introduces a novel training-free strategy, kNN-CLIP, which augments the model with a database of instance embeddings for semantic and panoptic segmentation that achieves zero forgetting and achieves state-of-the-art performance across large-vocabulary semantic and panoptic segmentation datasets.
- 29
- Model Selection for Two View Geometry:A Review29
This paper explores ways of automating the model selection process, with specific emphasis on the least squares problem, and the handling of implicit or nuisance parameters (which in this case equate to 3D structure).
- Revisiting Uncertainty Estimation and Calibration of Large Language Models28
The most comprehensive study to date of uncertainty estimation in LLMs, evaluating 80 models spanning open- and closed-source families, dense and Mixture-of-Experts (MoE) architectures, reasoning and non-reasoning modes, quantization variants and parameter scales, shows that LVU consistently outperforms TPU and NVU, offering stronger calibration and discrimination while being more interpretable.
- 28
- Learn what matters: cross-domain imitation learning with task-relevant embeddings28
This work proposes a scalable framework that enables cross-domain imitation learning without access to additional demonstrations or further domain knowledge, and demonstrates successful transfer of policies between considerably different domains, without extra supervision such as additional demonstrations, and in situations where other methods fail.
- 28
- NAPSAC: high noise, high dimensional model parameterisation - it's in the bag28
The NAPSAC (N Adjacent Points Sample Consensus) algorithm is derived and its performance in high dimensionality is shown to be superior to RANSAC on a hyperplane fitting problem.
- No Culture Left Behind: ArtELingo-28, a Benchmark of WikiArt with Captions in 28 Languages27
ArtELingo-28 is presented, a vision-language benchmark that spans 28 languages and encompasses approximately 200,000 annotations (140 annotations per image) and finds that cross-lingual transfer is more successful for culturally-related languages.
- 27
- FedBiP: Heterogeneous One-Shot Federated Learning with Personalized Latent Diffusion Models26
Federated Bi-Level Personalization (FedBiP) is proposed, which personalizes the pretrained LDM at both instance-level and concept-level and is the first approach to simultaneously address feature space heterogeneity and client data scarcity in OSFL.
- PoRF: Pose Residual Field for Accurate Neural Surface Reconstruction26
The pose residual field (PoRF) is introduced, a novel implicit representation that uses an MLP for regressing pose updates that is more robust than the conventional pose parameter optimisation due to parameter sharing that leverages global information over the entire sequence.
- 26
- AutoSimulate: (Quickly) Learning Synthetic Data Generation26
This work proposes an efficient alternative for optimal synthetic data generation, based on a novel differentiable approximation of the objective, which allows to optimize the simulator, which may be non-differentiable, requiring only one objective evaluation at each iteration with a little overhead.
- 24
- 24
- Object Proposal Generation Using Two-Stage Cascade SVMs24
This paper proposes a new object proposal generation method using two-stage cascade support vector machines (SVMs), where in the first stage linear filters are learned for predefined quantized scales/aspect-ratios independently, and in the second stage a global linear classifier is learned across all the quantized scales/aspect-ratios for calibration, so that all the windows from the first stage can be compared properly.
- 4D Gaussian Splatting: Modeling Dynamic Scenes with Native 4D Primitives23
This paper reformulates the reconstruction of a time-varying 3D scene as approximating its underlying spatiotemporal 4D volume by optimizing a collection of native 4D primitives, i.e., 4D Gaussians, with explicit geometry and appearance modeling.
- Rapid Adaptation in Online Continual Learning: Are We Evaluating It Right?23
This work proposes a novel metric for measuring adaptation of Online Continual Learning algorithms based on the accuracy on the near-future samples, where spurious correlations are removed, and finds that better generalization can be achieved by retaining and reusing past seen information.
- ReMaX: Relaxing for Better Training on Efficient Panoptic Segmentation23
This paper presents ReMaX that adds relaxation to mask predictions and class predictions during training for panoptic segmentation, and demonstrates that via these simple relaxation techniques during training, the model can be consistently improved by a clear margin without any extra computational cost on inference.
- DeformRS: Certifying Input Deformations with Randomized Smoothing23
This work reformulate certification in randomized smoothing setting for both general vector field and parameterized deformations and proposes DeformRS-VF and Deform RS-Par, respectively, which scales to large networks on large input datasets.
- 22
- 22
- Proximal Mean-Field for Neural Network Quantization22
This work designs an efficient iterative optimization procedure that involves stochastic gradient descent followed by a projection and proves that this simple projected gradient descent approach is, in fact, equivalent to a proximal version of the well-known mean-field method.
- 22
- Bipartite Graph Reasoning GANs for Person Pose and Facial Image Synthesis21
A novel bipartite graph reasoning Generative Adversarial Network (BiGraphGAN) that mitigates some of the challenges caused by pose deformation and proposes a new interaction-and-aggregation block to effectively update and enhance the feature representation capability of both a person’s shape and appearance in an interactive way.
- Joint Training of Generic CNN-CRF Models with Stochastic Optimization21
A new CNN-CRF end-to-end learning framework, which is based on joint stochastic optimization with respect to both Convolutional Neural Network and Conditional Random Field parameters, is proposed and empirically evaluates the method on the task of semantic labeling of body parts in depth images and shows that it compares favorably to competing techniques.
- OxfordTVG-HIC: Can Machine Make Humorous Captions from Images?20
This paper presents OxfordTVG-HIC (Humorous Image Captions), a large-scale dataset for humour generation and understanding and identifies the visual and linguistic cues influential for evoking humour prediction and generation and observes qualitatively that these cues are aligned with the benign violation theory of humour in cognitive psychology.
- Hierarchical Interaction Network for Video Object Segmentation from Referring Expressions20
This paper presents an end-to-end hierarchical interaction network (HINet) for the VOSRE problem, which leverages the feature pyramid produced by the visual encoder to generate multiple levels of multi-modal features and extracts signals of moving objects from optical input.
- Meta-Learning Deep Visual Words for Fast Video Object Segmentation20
This work develops a fast, causal algorithm that requires no finetuning, auxiliary inputs or post-processing, and segments a variable number of objects in a single forward-pass, achieving comparable accuracy to finetuned based methods, and state-of-the-art in terms of speed/accuracy trade-offs on four video segmentation datasets.
- 20
- Distributed Non-convex ADMM-based inference in large-scale random fields.20
This work proposes a parallel and distributed algorithm for solving discrete labeling problems in large scale random fields using a tree-based decomposition of the original optimization problem which is solved using a non convex form of the method of alternating direction method of multipliers (ADMM).
- Are LLMs Socially Adaptive? Contrasting Belief Evolution in Large Language Models and Humans19
The Belief-Reward Alignment Behavior Evolution Model (BREM) is introduced, a probabilistic framework that formalizes decision-making as a dynamic trade-off between maximizing extrinsic rewards and upholding intrinsic beliefs and finds that more advanced models better balance conflicting objectives by reducing belief action inconsistency.
- Deeply Explain CNN Via Hierarchical Decomposition19
A hierarchical decomposition framework to explain CNN’s decision-making process in a top-down manner with a gradient-based activation propagation module that can decompose any intermediate CNN decision to its lower layers and find the supporting features.
- MobileBrick: Building LEGO for 3D Reconstruction on Mobile Devices19
A novel multi-view RGBD dataset captured using a mobile device is introduced, which includes highly precise 3D ground-truth annotations for 153 object models featuring a diverse set of 3D structures.
- 19
- 19
- Layer Extraction with a Bayesian Model of Shapes19
An automatic 3D surface modelling system that extracts dense 3D surfaces from uncalibrated video sequences is described and a new segmentation method is described, which uses a specific prior model for layer shape.
- Deep Fully-Connected Part-Based Models for Human Pose Estimation18
A 2D multi-level appearance representation of the human body in RGB images, spatially modelled using a fully-connected graphical model based on a CNN body part detector, over which approximate inference is efficiently performed using the Mean-Field algorithm.
- Joint Object-Material Category Segmentation from Audio-Visual Cues18
This paper presents an approach that augments the available dense visual cues with sparse auditory cues in order to estimate dense object and material labels and demonstrates that this joint estimation of object andmaterial labels significantly outperforms the estimation of either category in isolation.
- Efficient Likelihood Learning of a Generic CNN-CRF Model for Semantic Segmentation.18
This work presents a CRF model were unary factors are dependent on a CNN, an efficient and scalable, maximum likelihood-based, learning procedure to infer all model parameters jointly and is the first to perform maximum likelihood learning for large-sized factor graphs with non-parametric potentials.
- Taxonomic Multi-class Prediction and Person Layout Using Efficient Structured Ranking18
An algorithm for structured output ranking that can be trained in a time linear in the number of samples under a mild assumption common to many computer vision problems: that the loss function can be discretized into a small number of values is proposed.
- Rethinking Safety in LLM Fine-tuning: An Optimization Perspective17
This work proposes a simple exponential moving average (EMA) momentum technique in parameter space that preserves safety performance by creating a stable optimization path and retains the original pre-trained model's safety properties, and proposes practical guidelines for maintaining both model performance and safety during adaptation.
- LLM Jailbreak Detection for (Almost) Free!17
A Free Jailbreak Detection (FJD) is proposed which prepends an affirmative instruction to the input and scales the logits by temperature to further distinguish between jailbreak and benign prompts through the confidence of the first token.
- 17
- Metric Attack and Defense for Person Re-identification17
This work proposes Adversarial Metric Attack, a parallel methodology to adversarial classification attacks, and presents an early attempt of training a metric-preserving network, thereby defending the metric against adversarial attacks.
- 17
- 17
- Unforgotten Safety: Preserving Safety Alignment of Large Language Models with Continual Learning16
This paper considers the fine-tuning-as-a-service setup where the user uploads their data to a service provider to get a customized model that excels on the user's selected task.
- Towards Interpretable Sequence Continuation: Analyzing Shared Circuits in Large Language Models16
This work identifies a key sub-circuit in both GPT-2 Small and Llama-2-7B responsible for detecting sequence members and for predicting the next member in a sequence and shows that this sub-circuit has effects on various math-related prompts.
- Dynamic Graph Message Passing Networks for Visual Recognition16
This paper proposes a dynamic graph message passing network, that significantly reduces the computational complexity compared to related works modelling a fully-connected graph, and outperforms fully- connected graphs while using substantially fewer floating-point operations and parameters.
- 16
- 16
- 16
- 16
- DreamPolisher: Towards High-Quality Text-to-3D Generation via Geometric Diffusion15
A two-stage Gaussian Splatting based approach that enforces geometric consistency among views, and uses a ControlNet driven refiner coupled with the geometric consistency term to improve both texture fidelity and overall consistency of the generated 3D asset.
- Rethinking Visual Prompting for Multimodal Large Language Models with External Knowledge15
This paper proposes a new visual prompt approach to integrate fine-grained external knowledge, gleaned from specialized vision models, into MLLMs, and can be effortlessly incorporated into various MLLMs, such as LLaVA and Mipha, considerably improving their visual understanding performance.
- Don’t FREAK Out: A Frequency-Inspired Approach to Detecting Backdoor Poisoned Samples in DNNs15
This work investigates the frequency sensitivity of Deep Neural Networks when presented with clean samples versus poisoned samples and proposes FREAK, a frequency-based poisoned sample detection algorithm that is simple yet effective against frequency backdoor attacks.
- 15
- A Projected Gradient Descent Method for CRF Inference Allowing End-to-End Training of Arbitrary Pairwise Potentials15
This paper develops a new inference and learning framework which can learn pairwise CRF potentials restricted only by their dependence on the image pixel values and the size of the support and empirically demonstrated that such learned potentials can improve segmentation accuracy.
- SETA: Scaling Environments for Terminal Agents14
This work constructs and releases SETA-Env, the largest open-source verifiable terminal RL dataset to date, containing over 4,500 environments, and demonstrates that SETA- Env provides high-quality training environments for terminal agents and serves as a valuable resource for advancing research on terminal-based agent learning.
- Memory-Driven Text-to-Image Generation14
Experimental results demonstrate that the proposed memory-driven semi-parametric approach to text-to-image generation produces more realistic images than purely parametric approaches, in terms of both visual fidelity and text-image semantic consistency.
- 14
- Scalable Cascade Inference for Semantic Image Segmentation14
This paper proposes performing cascaded inference that wraps around the α-expansion algorithm, which divides the large label set into smaller more manageable ones by way of a hierarchy, and dynamically subdivides the image into smaller and smaller regions during inference.
- Agentic World Modeling: Foundations, Capabilities, Laws, and Beyond13
The resulting roadmap connects previously isolated communities and charts a path from passive next-step prediction toward world models that can simulate, and ultimately reshape, the environments in which agents operate.
- 13
- Toward Robust Real-World Audio Deepfake Detection: Closing the Explainability Gap13
Novel explainability methods for state-of-the-art transformer-based audio deepfake detectors are introduced and a novel benchmark for real-world generalizability is open-source for real-world generalizability.
- 13
- 13
- Many-Shot from Low-Shot: Learning to Annotate Using Mixed Supervision for Object Detection13
This work introduces, for the first time, an online annotation module (OAM) that learns to generate a many-shot set of \emph{reliable} annotations from a larger volume of weakly labelled images, resulting in a fully end-to-end strategy that only requires a low- shot set of fully annotated images.
- 13
- Learning Disentangled Representations in Deep Generative Models13
This work introduces a new formulation of semi-supervised learning in variational autoencoders that permits flexible specification of probabilistic encoders as directed graphical models via a stochastic computation graph, containing both continuous and discrete latent variables, with conditional distributions parametrised by neural networks.
- Higher Order Priors for Joint Intrinsic Image, Objects, and Attributes Estimation13
A joint estimation of intrinsic scene properties recovered from an image, together with the estimation of the objects and attributes present in the scene is explored, able to capture the correlations between intrinsic properties, objects, and materials in a given scene.
- 12
- 12
- 12
- 12
- Traditional Classification Neural Networks are Good Generators: They are Competitive with DDPMs and GANs12
This work demonstrates that conventional neural network classifiers can generate high-quality images of a large number of categories, being comparable to the state-of-the-art generative models (e.g., DDPMs and GANs).
- On the Robustness of Quality Measures for GANs12
This work evaluates the robustness of quality measures of generative models such as Inception Score (IS) and Fr\'echet Inception Distance (FID) and shows that such metrics can also be manipulated by additive pixel perturbations.
- Deep Virtual Networks for Memory Efficient Inference of Multiple Tasks12
This work proposes a novel network architecture producing multiple networks of different configurations, termed deep virtual networks (DVNs), for different tasks, and shows that the proposed method outperforms existing approaches for multiple tasks.
- 12
- Efficient Lifelong Model Evaluation in an Era of Rapid Progress11
An efficient framework for model evaluation, Sort&Search (S&S), is introduced, which reuses previously evaluated models by leveraging dynamic programming algorithms to selectively rank and sub-select test samples and achieves highly-efficient approximate accuracy measurement.
- Estimation of Near-Instance-Level Attribute Bottleneck for Zero-Shot Learning11
A Near-Instance-Level Attribute Bottleneck (IAB) is introduced to alter class-level attribute vectors as well as visual features throughout the training phase to better reflect their naturalistic correspondences and corroborate the sustainability of this study as one of the most potential directions for ZSL.
- Semantic Score Distillation Sampling for Compositional Text-to-3D Generation11
This work introduces a novel SDS approach, Semantic Score Distillation Sampling (SemanticSDS), designed to effectively improve the expressiveness and accuracy of compositional text-to-3D generation, and unlocks the compositional capabilities of existing pre-trained diffusion models.
- 11
- Efficient Relaxations for Dense CRFs with Sparse Higher-Order Potentials11
It is shown that a dense CRF can be aggregated with sparse higher order potentials in a way that is amenable to continuous relaxations and solved a quadratic programming (QP) relaxation using the Frank-Wolfe algorithm and a linear programming (LP) relaxation by developing a proximal minimisation framework.
- 11
- 11
- Clustering Generative Adversarial Networks for Story Visualization10
The discriminator in the approach provides comprehensive feedback to the generator, regarding both image quality and story consistency at the same time, allowing to have a single-GAN-based network to produce high-quality synthetic results.
- 10
- Rethinking Semi-Supervised Learning in VAEs.10
A novel VAE model is developed, the reparameterized VAE (ReVAE), which "reparameterizes" supervision through auxiliary variables and a concomitant variational objective, and it is shown that the ReVAE can effectively learn meaningful representations of data.
- DGPose: Deep Generative Models for Human Body Analysis10
This work presents deep generative models for human body analysis in which the body pose and the visual appearance are disentangled, which allows independent manipulation of pose and appearance, and hence enables applications such as pose-transfer without specific training for such a task.
- 10
- 10
- Latent SVMs for Human Detection with a Locally Affine Deformation Field10
A latent deformable template model with a locally affine deformation field is proposed, which allows for more general and more natural deformations of the template while not over-fitting the data; and a novel inference method is provided for this kind of problem.
- Towards Agents That Know When They Don't Know: Uncertainty as a Control Signal for Structured Reasoning9
This work introduces an uncertainty-aware agent for query-conditioned multi-table summarization that leverages two complementary signals: retrieval uncertainty--entropy over multiple table-selection rollouts--and summary uncertainty--combining self-consistency and perplexity.
- LUNA: Language as Continuing Anchors for Referring Expression Comprehension9
This work proposes LUNA, which uses language as continuing anchors to guide box prediction in a Transformer decoder, and shows that language-guided location priors can be effectively exploited in aTransformer-based architecture.
- ROAM: A Rich Object Appearance Model with Application to Rotoscoping9
This work proposes a novel framework to capture and track the visual aspect of an arbitrary object in a scene, given a first closed outline of this object, that combines a collection of local foreground/background appearance models spread along the outline, a global appearance model of the enclosed object and a set of distinctive foreground landmarks.
- Coarse-to-fine Planar Regularization for Dense Monocular Depth Estimation9
This work proposes a framework for monocular SLAM, and its local model in particular, which optimizes simultaneously over depth and pose, and constrain the complexity of depth map updates, which provides a natural way to avoid poor local minima and reduces unknowns in the optimization.
- Detecting Multi-Agent Collusion Through Multi-Agent Interpretability8
It is suggested that model internals provide a complementary signal to text-level monitoring for detecting multi-agent collusion, and no single probing technique dominates across all collusion types, consistent with the framework's prediction that different anomaly types require different detection paradigms.
- Do Sparse Autoencoders Generalize? A Case Study of Answerability8
This work extensively evaluate SAE feature generalization across diverse, partly self-constructed answerability datasets for Gemma 2 SAEs, revealing that residual stream probes outperform SAE features within domains, but generalization performance differs sharply.
- Shh, don't say that! Domain Certification in LLMs8
This work introduces domain certification; a guarantee that accurately characterizes the out-of-domain behavior of language models that provides adversarial bounds as a certificate and proposes a simple yet effective approach, which is called VALID, that provides adversarial bounds as a certificate.
- TraceDet: Hallucination Detection from the Decoding Trace of Diffusion Large Language Models8
Extensive experiments on various open source D-LLMs demonstrate that TraceDet consistently improves hallucination detection, achieving an average gain in AUROC of 15.2% compared to baselines.
- Continual Learning on a Diet: Learning from Sparsely Labeled Streams Under Constrained Computation8
DietCL outperforms, by a large margin, all existing supervised CL algorithms as well as more recent continual semi-supervised methods, and meticulously allocates computational budget for both unlabeled and labeled data jointly.
- Focus On This, Not That! Steering LLMs with Adaptive Feature Specification8
FIT is introduced, which trains LLMs to condition their responses by focusing on specific features whilst ignoring others, leading to different behaviours based on what features are specified, and offers a lightweight, intrinsic mechanism for building more robust, fair, and easily controllable LLMs.
- Extracting Training Data from Document-Based VQA Models8
This work shows Vision-Language Models can memorize responses for training samples and regurgitate them even when the relevant visual information has been removed, indicating these models could divulge memorised sensitive information and therefore pose a privacy risk.
- Raising the Bar on the Evaluation of Out-of-Distribution Detection8
This paper defines 2 categories of OoD data using the subtly different concepts of perceptual/visual and semantic similarity to in-distribution (iD) data and proposes a GAN based framework for generating OoD samples from each of these 2 categories, given an iD dataset.
- 8
- 8
- Adversarial Metric Attack for Person Re-identification.8
This work proposes Adversarial Metric Attack, a parallel methodology to adversarial classification attacks, which can effectively generate adversarial examples for re-ID, and expects that this work can facilitate the development of robust feature learning with the experimental conclusions.
- 8
- 7
- 7
- Towards Certification of Uncertainty Calibration under Adversarial Attacks7
It is shown that attacks can significantly harm calibration, and thus certified calibration is proposed as worst-case bounds on calibration under adversarial perturbations.
- 7
- 7
- DGPose: Disentangled Semi-supervised Deep Generative Models for Human Body Analysis.7
A structured semi-supervised variational auto-encoder approach is adopted and present a deep generative model for human body analysis where the pose and appearance are disentangled in the latent space, allowing for pose estimation and hence enables applications such as pose-transfer without being explicitly trained for such a task.
- A shape hierarchy for 3D modelling from video7
The method provides a modelling process which requires the user to provide only high level scene information, the remaining detail being provided through geometric analysis of the image set, and allows a non-expert user to rapidly and intuitively create a visually convincing 3D model of a real world scene from an image set.
- LLM Agents Are the Antidote to Walled Gardens6
The position is that the ML community should embrace the ability for any two digital services to exchange data seamlessly using AI-mediated adapters while building the appropriate frameworks to mitigate the downsides.
- MatchDiffusion: Training-Free Generation of Match-Cuts6
This work presents the first training-free method for match-cut generation using text-to-video diffusion models, and leverages a key property of diffusion models: early denoising steps define the scene's broad structure, while later steps add details.
- AI Deception: Risks, Dynamics, and Controls6
This project provides a comprehensive and up-to-date overview of the AI deception field, covering its core concepts, methodologies, genesis, and potential mitigations, and organizes the landscape of AI deception research as a deception cycle, consisting of two key components: deception emergence and deception treatment.
- Responsible AI Governance: A Response to UN Interim Report on Governing AI for Humanity6
The response highlights opportunities for promoting equitable, secure, and inclusive AI ecosystems, which should be supported by investments in infrastructure and multi-stakeholder collaborations across jurisdictions, and underscores challenges, including societal inequalities exacerbated by AI, ethical concerns, and environmental impacts.
- 6
- Class-Agnostic Segmentation Loss and Its Application to Salient Object Detection and Segmentation6
This paper presents a novel loss function, called class-agnostic segmentation (CAS) loss, which doesn’t require to define the label of a class a-priori, rather the CAS loss clusters regions with similar appearance together in a weakly-supervised manner.
- 6
- Revisiting Deep Structured Models for Pixel-Level Labeling with Gradient-Based Inference6
A new inference and learning framework which can learn arbitrary pairwise conditional random field (CRF) potentials which can improve segmentation accuracy and prove that certain label-class interactions are indeed better modeled by a non-Gaussian potential.
- Select-then-Solve: Paradigm Routing as Inference-Time Optimization for LLM Agents5
This work argues that reasoning paradigm selection should be a per-task decision made by a learned router, not a fixed architectural choice, and proposes a select-then-solve approach.
- The Alignment Curse: Modality Alignment Supercharges Audio Attacks via Text Transfer5
This work analyzes the connection between modality alignment and cross-modality jailbreak transfer, and demonstrates strong cross-model transferability and shows that text-transferred audio attacks remain effective even under a stricter audio-only access threat model.
- AI-Driven Automation Can Become the Foundation of Next-Era Science of Science Research5
A preliminary multi-agent system is presented as an illustrative example to simulate research societies, showcasing AI's ability to replicate real-world research patterns and accelerate progress in Science of Science research.
- Interpreting Learned Feedback Patterns in Large Language Models5
It is hypothesize that LLMs with LFPs accurately aligned to the fine-tuning feedback exhibit consistent activation patterns for outputs that would have received similar feedback during RLHF.
- Saliency-guided meta-hallucinator for few-shot learning5
This work proposes a novel saliency-based end-to-end meta-hallucinator, where a saliency detector produces foregrounds and backgrounds of support images, and is the first to leverage saliency detection for few-shot learning.
- Segment, Select, Correct: A Framework for Weakly-Supervised Referring Segmentation5
A novel weakly-supervised framework that tackles RIS by decomposing it into three steps: obtaining instance masks for the object mentioned in the referencing instruction (segment), using zero-shot learning to select a potentially correct mask for the given instruction (select), and bootstrapping a model which allows for fixing the mistakes of zero-shot selection (correct).
- 5
- Improving Few-shot Learning by Spatially-aware Matching and CrossTransformer5
This paper proposes a novel Spatially-aware Matching (SM) scheme to effectively perform matching across multiple scales and locations, and learn image relations by giving the highest weights to the best matching pairs.
- MP25
A momentum contrast framework that combines pointwise and pairwise learning for recommendation and finds the soft-labeling property of pairwise labels could be utilized to alleviate the bias of pointwise labels.
- 5
- 5
- 5
- 5
- 5
- 5
- LongCoT: Benchmarking Long-Horizon Chain-of-Thought Reasoning4
LongCoT is introduced, a scalable benchmark of 2,500 expert-designed problems spanning chemistry, mathematics, computer science, chess, and logic to isolate and directly measure the long-horizon CoT reasoning capabilities of frontier models.
- 4
- NeRF-VPT: Learning Novel View Representations with Neural Radiance Fields via View Prompt Tuning4
It is substantiated that the NeRF-VPT significantly elevates baseline performance and proficiently generates more high-quality novel view images than all the compared state-of-the-art methods.
- Select to Perfect: Imitating desired behavior from large multi-agent data4
The concept of an agent's Exchange Value, which quantifies an individual agent's contribution to the collective desirability score, is proposed, which allows us to selectively imitate agents with a positive effect, e.g., only imitating agents that are unlikely to cause incidents.
- Measuring Value Alignment4
A novel formalism to quantify the alignment between AI systems and human values is introduced, using Markov Decision Processes (MDPs) as the foundational model, and the concept of values as desirable goals tied to actions and norms as behavioral guidelines is explored.
- 4
- 4
- 4
- 4
- Real-time depth processing for embedded platforms4
This work demonstrates a computationally and memory efficient implementation of a stereo block-matching algorithm in FPGA that achieves a throughput of 577 fps at standard VGA resolution whilst consuming less than 3 Watts of power.
- 4
- Distinguishable Deletion: Unifying Knowledge Erasure and Refusal for Large Language Model Unlearning3
- In Which Areas of Technical AI Safety Could Geopolitical Rivals Cooperate?3
This analysis finds that research into AI verification mechanisms and shared protocols may be suitable areas for international cooperation, and aims to help researchers and governments identify and mitigate the risks of international cooperation on AI safety research, so that the benefits of cooperation can be fully realised.
- 3
- 3
- 3
- 3
- Illusory Attacks: Information-Theoretic Detectability Matters in Adversarial Attacks3
It is concluded that agents must seek realism feedback to be robust to illusionary attacks, a novel form of attack that is consistent with the world model of the victim agent.
- Breaking Shortcut: Exploring Fully Convolutional Cycle-Consistency for Video Correspondence Learning3
- 3
- Few-shot Action Recognition via Improved Attention with Self-supervision3
This paper tackles a more challenging task of few-shot action-recognition from video clips with a simple framework which is both flexible and easy to implement and compares favorably with several state-of-the-art baselines on HMDB51, miniMIT and UCF101 datasets.
- 3
- Learning Arbitrary Pairwise Potentials in CRFs for Semantic Segmentation3
A new inference and learning framework which can learn arbitrary pairwise CRF potentials which can improve segmentation accuracy and that certain label class interactions are indeed better modelled by a non-Gaussian potential is developed.
- Probabilistic Active Recognition of Multiple Objects Using Hough-Based Geometric Matching Features3
This work states that 3D Object recognition is an important task for mobile platforms to dynamically interact in human environments and plays a fundamental role in the areas of automated surveillance, Simultaneous Localization and Mapping applications for robots and video retrieval.
- Probabilistic models for 2D active shape recognition using Fourier descriptors and mutual information3
To improve upon initial shape recognition results, multinomial and Gaussian probabilistic models are built using the extracted Fourier descriptors and it is shown how actively looking for cues using mutual information can improve the overall results.
- Estimating uncertainty in dense stereo disparity maps3
A new, linear-time, exact method for recovering entire distributions for disparity at all pixels is presented, using a recent extension of the forward-backward algorithm to work with two unsynchronised input streams, rather than just one as in the conventional case.
- ELAN4D: Embodiment-Centric 4D Supervision for Vision-Language-Action Models via Plug-and-Play Adaptation2
ELAN4D is proposed, an embodiment-centric, 4D-aware training framework that enhances VLA policies with future robot keypoint tracks as predictive spatio-temporal supervision and consistently improves over strong VLA baselines, achieving the best overall performance and substantial gains under out-of-distribution perturbations.
- Improving Semantic Uncertainty Quantification in Language Model Question-Answering via Token-Level Temperature Scaling2
This work demonstrates that optimising a single scalar temperature, which, it is argued, provides a suitable inductive bias, is a surprisingly simple yet effective solution to semantic calibration and discrimination.
- When Vision Meets Graphs: A Survey on Graph Reasoning and Learning2
This survey provides a first systematic overview of the emerging area of vision meets graphs, which treats visual depictions of graphs as first-class inputs for reasoning and learning, and organize existing work into three threads.
- 2
- 2
- SAGE: Scalable Ground Truth Evaluations for Large Sparse Autoencoders2
SAGE: Scalable Autoencoder Ground-truth Evaluation, a ground truth evaluation framework for SAEs that scales to large state-of-the-art SAEs and models that paves the way for generalizable, large-scale evaluations of SAEs in interpretability research.
- Self-consistent Validation for Machine Learning Electronic Structure2
This method integrates machine learning with self-consistent field methods to achieve both low validation cost and interpret-ability and enables exploration of the model's ability with active learning and instills confidence in its integration into real-world studies.
- Specify and Edit: Overcoming Ambiguity in Text-Based Image Editing2
This work uses a large language model (LLM) to decompose the input instruction into specific instructions, i.e. well-defined interventions to apply to the input image to satisfy the user's request.
- Exploring Non-additive Randomness on ViT against Query-Based Black-Box Attacks2
This work first taxonomize the stochastic defense strategies against QBBA, and proposes to explore non-additive randomness in models to defend against it, focusing on underexplored Vision Transformers based on their flexible architectures.
- 2
- 2
- 2
- 2
- 2
- Rethinking Class Relations: Absolute-relative Few-shot Learning.2
A novel absolute-relative learning paradigm is proposed to fully take advantage of label information to refine the image representations and correct the relation understanding and improves the performance of several the state-of-the-art models on publicly available datasets.
- 2
- 2
- 2
- 2
- 2
- The Path Matters: Learning a Token-Commitment Policy for Diffusion Language Models1
Experiments show that TraceLock improves the quality-step tradeoff over heuristic and learned baselines, with particularly stable behavior under cross-setting deployment, suggesting that frozen diffusion language models expose a learnable space of commitment trajectories beyond confidence-based decoding.
- 1
- 1
- Safe Autoregressive Image Generation with Iterative Self-Improving Codebooks1
This work uses the understanding and judgment capabilities of the unified multimodal model itself to identify unsafe generated images without human annotation, and proposes iterative self-improving codebooks for safe autoregressive generation.
- Advances and Innovations in the Multi-Agent Robotic System (MARS) Challenge1
The Multi-Agent Robotic System (MARS) Challenge is proposed, held at the NeurIPS 2025 Workshop on SpaVLE, where participants explore multi-agent embodied planning using vision-language models (VLMs) to coordinate tasks and policy execution to perform robotic manipulation in dynamic environments.
- Ego to World: Collaborative Spatial Reasoning in Embodied Systems via Reinforcement Learning1
The Ego-to-World (E2W) benchmark is introduced, which evaluates a vision-language model's ability to fuse heterogeneous viewpoints across three tasks: (i) global counting, (ii) relational location reasoning, and (iii) action-oriented grasping that requires predicting view-specific image coordinates.
- Retrieval-Augmented Linguistic Calibration1
Retrieval-Augmented Linguistic Calibration (RALC), a lightweight post-hoc pipeline that propagates calibrated confidence signals back into natural language via retrieval-augmented rewriting and introduces faithfulness as a complementary evaluation dimension and presents Faithfulness Divergence (FD), an information-theoretic metric quantifying the surprise induced in audience beliefs upon truth revelation.
- NeurIPS 2025 E2LM Competition : Early Training Evaluation of Language Models1
This competition tackles the challenge of designing scientific knowledge evaluation tasks specifically tailored for measuring early training progress of language models, and seeks to make foundational LLM research more systematic and benchmark-informed from the earliest phases of model development.
- MATRIX: Multimodal Agent Tuning for Robust Tool-Use Reasoning1
A vision-centric agent tuning framework that automatically synthesizes multimodal trajectories, generates step-wise preference pairs, and trains a VLM controller for robust tool-use reasoning, demonstrating scalable and effective multimodal tool use.
- 1
- 1
- 1
- Deconstructed Generation-Based Zero-Shot Model1
This paper begins by formally defining semantic generalization, then looks into approaches for reducing the semantic weak generalization problem and minimizing its negative influence on classifier training.
- 1
- 1
- 1
- 1
- 1
- 1
- Knowing who to listen to: Prioritizing experts from a diverse ensemble for attribute personalization1
- 1
- Global Stereo Reconstruction under1
This paper shows that inference with triple cliques can be effectively performed, and develops a development of recent extensions to � -expansion, based on the "QPBO" algorithm.
- –
- –
- –
- SciOrch: Learning to Orchestrate Expert LLMs for Solving Frontier Multimodal Scientific Reasoning Tasks–
SciOrch is presented, a framework that trains a lightweight 8B model to orchestrate frontier LLMs for scientific reasoning, and attains the best accuracy on both SGI and SFE with less than half the API cost of typical multi-agent methods.
- Technical Report on the CVPR 2026@AdvML Workshop Challenge–
The CVPR 2026@AdvML Workshop Challenge on adversarial multimodal attacks against autonomous-driving VLAs is presented, providing a practical reference for future robustness evaluation and defense design in multimodal autonomous-driving systems.
- –
- –
- –
- Data Journalist Agent: Transforming Data into Verifiable Multimodal Stories–
This work introduces Data Journalist Agent (Data2Story), a multi-agent framework that orchestrates specialized roles into a single virtual newsroom, enabling more evidence-based, transparent, and verifiable reporting.
- –
- –
- –
- –
- $D^2$-Monitor: Dynamic Safety Monitoring for Diffusion LLMs via Hesitation Signals–
This work proposes D 2 -Monitor, a bi-level safety monitor for D-LLMs, which achieves state-of-the-art performance with a compact parameter footprint, and exhibits the best trade-off between effectiveness and efficiency relative to 8 baselines.
- –
- –
- –
- –
- –
- –
- –
- –
- Plain Transformers Can be Powerful Graph Learners–
Three simple modifications to the plain Transformer are proposed to render it applicable to graphs without introducing major architectural distortions, and empirical evaluation on the expressiveness benchmark reveals noteworthy realized expressiveness in the graph isomorphism.
- Sensitivity Analysis for Climate Science with Generative Flow Models–
This work applies the adjoint state method for calculating gradients in generative flow models to the cBottle generative model, trained on ERA5 and ICON data, and quantitatively validate the computed sensitivities against the model's own outputs.
- DEEDEE: Fast and Scalable Out-of-Distribution Dynamics Detection–
The results indicate that diverse anomaly types often imprint on RL trajectories through a small set of low-order statistics, suggesting a compact foundation for OOD detection in complex environments.
- –
- –
- –
- –
- –
- Revealing Decurve Flows for Generalized Graph Propagation–
This study addresses the limitations of the traditional analysis of message-passing, central to graph learning, by defining Generalized Propagation Neural Networks (GPNNs), a framework that unifies most propagation-based graph neural networks.
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- Implicit Communication as Minimum Entropy Coupling.–
This work isolates the implicit communication problem by identifying a class of partially observable common-payoff games, which it is shown can be attributed to implicit communication, and introduces a principled method based on minimum entropy coupling that leverages the structure of implicit referential games, yielding a new perspective on implicit communication.
- –
- Scalable FPGA Median Filtering using Multiple Efficient Passes–
This work proposes a novel way of substituting a 2-D median filter on an FPGA with a sequence of directional median filters, in this case in the pursuit of anFPGA implementation that achieves better scalability and hardware efficiency without sacrificing accuracy.
- –
- –
- –
- –
- –
- –
- –
- –
- Priors for new view synthesis–
This thesis presents a meta-analyses of the determinants of infectious disease in eight operation theatres of the central nervous system and shows clear patterns of disease progression that are consistent with those of previous studies of immune-inflammatory bowel disease.
Publication data from OpenAlex, with missing venues and authors filled in from Crossref; citation counts are the higher of OpenAlex and Semantic Scholar, last synced 2026-10-12. One-sentence summaries under some papers are written by Semantic Scholar’s model. Citation counts may be lower than on Google Scholar, which indexes more sources.
Report an error
Wrong papers, two people merged into one, or a profile that should not be here? Tell us and we will fix or hide it.