Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileAcademic lineage
Possible advisorsa guess from early papers, not confirmed
- Qi TianSuggested from co-authorship
Is this you? Claim this profile to confirm or dismiss it.
Works216 from public data
- 4,660
- Random Erasing Data Augmentation4,342
In training, Random Erasing randomly selects a rectangle region in an image and erases its pixels with random values and yields consistent improvement over strong baselines in image classification, object detection and person re-identification.
- Beyond Part Models: Person Retrieval with Refined Part Pooling (and A Strong Convolutional Baseline)2,522
This paper targets at learning discriminative part-informed features for person retrieval and lays emphasis on the content consistency within each part, resulting in refined parts with enhanced within-part consistency.
- Beyond Part Models: Person Retrieval with Refined Part Pooling.2,442
This paper targets at learning discriminative part-informed features for person retrieval and lays emphasis on the content consistency within each part, resulting in refined parts with enhanced within-part consistency.
- Unlabeled Samples Generated by GAN Improve the Person Re-identification Baseline in Vitro2,076
A simple semisupervised pipeline that only uses the original training set without collecting extra data, which effectively improves the discriminative ability of learned CNN embeddings and proposes the label smoothing regularization for outliers (LSRO).
- Re-ranking Person Re-identification with k-Reciprocal Encoding1,681
This paper proposes a k-reciprocal encoding method to re-rank the re-ID results, and hypothesis is that if a gallery image is similar to the probe in the k- Reciprocal nearest neighbors, it is more likely to be a true match.
- Towards Real-Time Multi-Object Tracking1,148
This work incorporates the appearance embedding model into a single-shot detector, such that the model can simultaneously output detections and the corresponding embeddings, and is formulated as a multi-task learning problem.
- MARS: A Video Benchmark for Large-Scale Person Re-Identification1,088
It is shown that CNN in classification mode can be trained from scratch using the consecutive bounding boxes of each identity, and the learned CNN embedding outperforms other competing methods considerably and has good generalization ability on other video re-id datasets upon fine-tuning.
- Circle Loss: A Unified Perspective of Pair Similarity Optimization1,086
The Circle loss is demonstrated, which has a unified formula for two elemental deep feature learning paradigms, learning with class-level labels and pair-wise labels, and the superiority of the Circle loss on a variety ofDeep feature learning tasks.
- Image-Image Domain Adaptation with Preserved Self-Similarity and Domain-Dissimilarity for Person Re-identification1,011
Through domain adaptation experiment, it is shown that images generated by SPGAN are more suitable for domain adaptation and yield consistent and competitive re-ID accuracy on two large-scale datasets.
- A Discriminatively Learned CNN Embedding for Person Reidentification964
This article proposes a Siamese network that simultaneously computes the identification loss and verification loss and learns a discriminative embedding and a similarity measurement at the same time, thus taking full usage of the re-ID annotations.
- 922
- Improving person re-identification by attribute and identity learning888
An attribute-person recognition (APR) network is proposed, a multi-task network which learns a re-ID embedding and at the same time predicts pedestrian attributes, and demonstrates that by learning a more discriminative representation, APR achieves competitive re-IDs performance compared with the state-of-the-art methods.
- Joint Discriminative and Generative Learning for Person Re-Identification856
This paper proposes a joint learning framework that couples re-id learning and data generation end-to-end and renders significant improvement over the baseline without using generated data, leading to the state-of-the-art performance on several benchmark datasets.
- Person Re-identification in the Wild830
A new dataset, PRW, is introduced to evaluate Person Re-identification in the Wild, and it is shown that pedestrian detection aids re-ID through two simple yet effective improvements: a cascaded fine-tuning strategy that trains a detection model first and then the classification model, and a Confidence Weighted Similarity (CWS) metric that incorporates detection scores into similarity measurement.
- Taking a Closer Look at Domain Shift: Category-Level Adversaries for Semantics Consistent Domain Adaptation829
A category-level adversarial network is introduced, aiming to enforce local semantic consistency during the trend of global alignment, to take a close look at the category- level data distribution and align each class with an adaptive adversarial loss.
- SVDNet for Pedestrian Retrieval825
This paper proposes the SVDNet for retrieval problems, with focus on the application of person re-identification (reID), and shows that RRI effectively reduces the correlation among the projection vectors, produces more discriminative FC descriptors, and significantly improves the re-ID accuracy.
- SIFT Meets CNN: A Decade Survey of Instance Retrieval756
A comprehensive survey of instance retrieval over the last decade, presenting milestones in modern instance retrieval, reviews a broad selection of previous works in different categories, and provides insights on the connection between SIFT and CNN-based methods.
- Invariance Matters: Exemplar Memory for Domain Adaptive Person Re-Identification685
Results on three re-ID domains show that the domain adaptation accuracy outperforms the state of the art by a large margin and the three invariance properties and the proposed memory are indispensable towards an effective domain adaptation system.
- Unsupervised Person Re-identification636
A progressive unsupervised learning (PUL) method to transfer pretrained deep representations to unseen domains and demonstrates that PUL outputs discriminative features that improve the re-ID accuracy.
- Camera Style Adaptation for Person Re-identification635
CamStyle can serve as a data augmentation approach that smooths the camera style disparities in person re-identification and demonstrate consistent improvement in all systems regardless of the extent of over-fitting.
- Dual-path Convolutional Image-Text Embeddings with Instance Loss617
An end-to-end dual-path convolutional network to learn the image and text representations based on an unsupervised assumption that each image/text group can be viewed as a class, which allows the system to directly learn from the data and fully utilize the supervision.
- Pose-Invariant Embedding for Deep Person Re-Identification598
This paper introduces pose-invariant embedding (PIE) as a pedestrian descriptor and shows that PoseBox alone yields decent re-ID accuracy and that when integrated in the PBF network, the learned PIE descriptor produces competitive performance compared with state-of-the-art approaches.
- A Bottom-Up Clustering Approach to Unsupervised Person Re-Identification589
The experimental results demonstrate that the bottom-up clustering approach to jointly optimize a convolutional neural network and the relationship among the individual samples is not only superior to state-of-the-art unsupervised re-ID approaches, but also performs favorably than competing transfer learning and semi-supervised learning methods.
- Pedestrian Alignment Network for Large-scale Person Re-Identification519
This work introduces the pedestrian alignment network (PAN) which allows discriminative embedding learning pedestrian alignment without extra annotations, and observes that when the convolutional neural network learns to discriminate between different identities, the learned feature maps usually exhibit strong activations on the human body rather than the background.
- Generalizing a Person Retrieval Model Hetero- and Homogeneously498
A Hetero-Homogeneous Learning (HHL) method is introduced that enforces two properties simultaneously: camera invariance, learned via positive pairs formed by unlabeled target images and their camera style transferred counterparts and domain connectedness, implemented by homogeneous learning.
- 317
- 303
- Dissecting Person Re-Identification From the Viewpoint of Viewpoint282
A large-scale synthetic data engine, PersonX, composed of hand-crafted 3D person models, which is able to synthesize pedestrians by setting the visual variables to arbitrary values, and quantitatively analyzes the influence of pedestrian rotation angle on re-ID accuracy.
- REPA-E: Unlocking VAE for End-to-End Tuning with Latent Diffusion Transformers242
It is shown that while diffusion loss is ineffective, end-to-end training can be unlocked through the representation-alignment (REPA) loss - allowing both VAE and diffusion model to be jointly tuned during the training process.
- 226
- Packing and Padding: Coupled Multi-index for Accurate Image Retrieval223
A coupled Multi-Index (c-MI) framework to perform feature fusion at indexing level, which improves the retrieval accuracy significantly, while consuming only half of the query time compared to the baseline, and is well complementary to many prior techniques.
- Learning to Adapt Invariance in Memory for Person Re-identification204
This work considers the problem of unsupervised domain adaptation in person re-identification (re-ID), which aims to transfer knowledge from the source domain to the target domain, and proposes a novel adaptation framework w.r.t three types of underlying invariance, i.e., Exemplar-Invariance, Camera-Invarance, and Neighborhood- Invariance.
- Linkage Based Face Clustering via Graph Convolution Network200
This paper presents an accurate and scalable approach to the face clustering task, and shows that the proposed method does not need the number of clusters as prior, is aware of noises and outliers, and can be extended to a multi-view version for more accurate clustering accuracy.
- Dynamic Label Graph Matching for Unsupervised Video Re-identification195
A dynamic graph matching (DGM) method is proposed that iteratively updates the image graph and the label estimation process by learning a better feature space with intermediate estimated labels and is robust to noisy initial training data.
- Learning Object Relation Graph and Tentative Policy for Visual Navigation178
Three complementary techniques, object relation graph (ORG), trial-driven imitation learning (IL), and a memory-augmented tentative policy network (TPN), which improves visual representation learning by integrating object relationships, including category closeness and spatial correlations are proposed.
- Few-Example Object Detection with Model Communication167
This paper studies object detection using a large pool of unlabeled images and only a few labeled images per category, named “few-example object detection”, and embeds multiple detection models in this framework, which has proven to outperform the single model baseline and the model ensemble method.
- Multiview Detection with Feature Perspective Transformation166
This work proposes a novel multiview detection system, MVDet, which takes an anchor-free approach to aggregateMultiview information by projecting feature maps onto the ground plane (bird's eye view) and applies large kernel convolutions on theGround plane feature map to resolve any remaining spatial ambiguity.
- Few-shot Object Detection165
This paper studies object detection using a large pool of unlabeled images and only a few labeled images per category, named “few-example object detection”, and embeds multiple detection models in this framework, which has proven to outperform the single model baseline and the model ensemble method.
- 159
- Macro-Micro Adversarial Network for Human Parsing153
Compared with traditional adversarial networks, the proposed Macro-Micro Adversarial Net not only enforces local and semantic consistency explicitly, but also avoids the poor convergence problem of adversarial Networks when handling high resolution images.
- 152
- Positive Sample Propagation along the Audio-Visual Event Line149
A new positive sample propagation (PSP) module is proposed to discover and exploit the closely related audio-visual pairs by evaluating the relationship within every possible pair by constructing an all-pair similarity map between each audio and visual segment.
- Simulating Content Consistent Vehicle Datasets with Attribute Descent149
This paper introduces a large-scale synthetic dataset VehicleX, which contains 1,362 vehicles of various 3D models with fully editable attributes, and proposes an attribute descent approach to let VehicleX approximate the attributes in real-world datasets.
- 147
- Are Labels Always Necessary for Classifier Accuracy Evaluation?145
This work constructs a meta-dataset: a dataset comprised of datasets generated from the original images via various transformations such as rotation, background substitution, foreground scaling, etc, and reports a reasonable and promising prediction of the model accuracy.
- 145
- 144
- Thorax disease classification with attention guided convolutional neural network134
This paper proposes to integrate the global and local cues into a three-branch attention guided convolution neural network (AG-CNN) to identify thorax diseases and demonstrates that after integrating the local cues with the global information, the average AUC scores are improved by AG-CNN.
- 131
- Deep Adversarial Attention Alignment for Unsupervised Domain Adaptation: The Benefit of Target Expectation Maximization127
This paper proposes an attention alignment scheme on all the target convolutional layers to uncover the knowledge shared by the source domain, and estimates the posterior label distribution of the unlabeled data for target network training.
- Deep learning corrosion detection with confidence122
This paper presents a deep learning corrosion detector that performs pixel-level segmentation of corrosion and three Bayesian variants are presented that provide uncertainty estimates depicting the confidence levels at each pixel, to better inform decision makers.
- 121
- CycAs: Self-supervised Cycle Association for Learning Re-identifiable Descriptions106
A different unsupervised method is introduced that allows us to learn pedestrian embeddings from raw videos, without resorting to pseudo labels, and adapt the discrete association process to a differentiable form, such that end-to-end training becomes feasible.
- The 4th AI City Challenge102
The 4th annual edition of the AI City Challenge has attracted 315 participating teams, who leverage city-scale real traffic data and high-quality synthetic data to compete in four challenge tracks, and results show promise that AI technology can enable smarter and safer transportation systems.
- The 5th AI City Challenge101
The fifth AI City Challenge attracted 305 participating teams across 38 countries, who leveraged city-scale real traffic data and high-quality synthetic data to compete in five challenge tracks, showing the promise of AI in Smarter Transportation.
- What matters for Representation Alignment: Global Information or Spatial Structure?97
Surprisingly, the simple method, termed iREPA, consistently improves convergence speed of REPA, across a diverse set of vision encoders, model sizes, and training variants (such as REPA, REPA-E, Meanflow, JiT etc).
- 95
- Multiview Detection with Shadow Transformer (and View-Coherent Data Augmentation)93
This paper proposes a novel multiview detector, MVDeTr, that adopts a newly introduced shadow transformer to aggregate multiv view information, and proposes an effective training scheme that includes a new view-coherent data augmentation method, which applies random augmentations while maintaining multivview consistency.
- Accurate Image Search with Multi-Scale Contextual Evidences92
This paper proposes to integrate discriminative cues from multiple contextual levels, i.e., local, regional, and global, via probabilistic analysis, and shows that CNN feature is complementary to SIFT due to its semantic awareness and compares favorably to several other descriptors such as GIST, HSV, etc.
- 92
- Attention-based Pyramid Aggregation Network for Visual Place Recognition90
An Attention-based Pyramid Aggregation Network (APANet) is presented, which is trained in an end-to-end manner for place recognition and significantly improves the widely used PCA whitening by reasonably limiting the impact of over-counting.
- Part-Based Deep Hashing for Large-Scale Person Re-Identification90
This paper makes the attempt in integrating deep learning and hashing into one framework to evaluate the efficiency and accuracy for large-scale person re-id, and proposes Part-based Deep Hashing (PDH), in which batches of triplet samples are employed as the input of the deep hashing architecture.
- Aesthetic Post-Training Diffusion Models from Generic Preferences with Step-by-step Preference Optimization88
Step-by-step preference optimization (SPO) is introduced that discards the propagation strategy and allows fine-grained image details to be assessed and finds that aesthetics can be significantly enhanced by accumulating these improved minor differences.
- 87
- In Search of Lost Online Test-Time Adaptation: A Survey82
These findings diverge from existing literature, revealing that transformers demonstrate heightened resilience to diverse domain shifts, and the efficacy of many OTTA methods relies on large batch sizes, and stability in optimization and resistance to perturbations are crucial during adaptation, particularly when the batch size is 1.
- 82
- Bayes Merging of Multiple Vocabularies for Scalable Image Retrieval82
This paper proposes a Bayes merging approach to down-weight the indexed features in the intersection set through explicitly modeling the correlation problem in a probabilistic view, and a joint similarity on both image- and feature-level is estimated for the indexed Features in the intersections set.
- Dual-Path Convolutional Image-Text Embedding.77
This paper builds a convolutional network amenable for fine-tuning the visual and textual representations, where the entire network only contains four components, i.e., convolution layer, pooling layer, rectified linear unit function (ReLU), and batch normalisation.
- Correlating Edge, Pose With Parsing76
This paper proposes a CorrPM employing a heterogeneous non-local block to discover the spatial affinity among feature maps from the edge, pose and parsing and finds that uncovering the correlation among the three factors is a superior way of leveraging the pivotal contextual cues provided by edges and poses.
- Twitter100k: A Real-World Dataset for Weakly Supervised Cross-Media Retrieval71
Inspired by the characteristic of Twitter100k, a method to integrate optical character recognition into cross-media retrieval is proposed and the experiment results show that the proposed method improves the baseline performance.
- 69
- Visualizing Adapted Knowledge in Domain Transfer69
This work proposes a source-free image translation method that generates source-style images using only target images and the two models to depict the adapted knowledge and finds that generated images successfully capture the style difference between the two domains.
- The 8th AI City Challenge63
The eighth AI City Challenge highlighted the convergence of computer vision and artificial intelligence in areas like retail, warehouse settings, and Intelligent Traffic Systems, presenting significant research opportunities, with participants setting new benchmarks.
- High-Fidelity Guided Image Synthesis with Latent Diffusion Models61
A novel guided image synthesis framework is proposed, which addresses this problem by modelling the output image as the solution of a constrained optimization problem and shows that while computing an exact solution to the optimization is infeasible, an approximation can be achieved while just requiring a single pass of the reverse diffusion process.
- 59
- 58
- 56
- 55
- 55
- InterActive: Inter-Layer Activeness Propagation51
InterActive is presented, a novel algorithm which computes the activeness of neurons and network connections, and achieves state-of-the-art classification performance on a wide range of image datasets.
- Improving Driver Gaze Prediction With Reinforced Attention48
The reinforced attention (RA) model is proposed as a regulatory mechanism to increase prediction density, built directly on top of existing methods, making it complementary to current approaches.
- 47
- 42
- Pose-Based View Synthesis for Vehicles: A Perspective Aware Method41
This paper proposes a perspective-aware generative model (PAGM) that can generate high-quality vehicle images with fine details and is superior to several competing approaches employing either GAN or the perspective transformation.
- 40
- 40
- Modality-Invariant Image-Text Embedding for Image-Sentence Matching38
A modality-invariant image-text embedding for image-sentence matching is proposed by involving adversarial learning, and a modality classification network is designed with an adversarial loss, which classifies an embedding into either the image or text modality.
- 36
- Exploiting Hierarchical Activations of Neural Network for Image Retrieval36
A Multi-layer Orderless Fusion (MOF) approach to integrate the activations of CNN in the Bag-of-Words (BoW) framework is proposed, through only one forward pass in the network, to extract multi-layer CNN activation of local patches.
- Perceive, Reflect, and Plan: Designing LLM Agent for Goal-Directed City Navigation without Instructions32
A novel agentic workflow featured by LLaVA-7B can be fine-tuned to perceive the direction and distance of landmarks with sufficient accuracy for city navigation and reflection is achieved through a memory mechanism, which can be retrieved with current perception for effective decision argumentation.
- Vehicle Re-Identification with Location and Time Stamps31
This paper introduces an ensemble technique to combine the informative cues of multiple Re-ID models effectively and builds up a system to acquire the vehicle location and time stamps.
- Learning completed discriminative local features for texture classification31
The proposed CDLF improves the distinctiveness of LBP-like feature vectors in two aspects: in the pattern encoding stage, it learns a transformation matrix using labeled data, which significantly increases the discrimination power of the encoded binary strings; in the histogram accumulation step, it uses an adaptive weight strategy to consider the contributions of pixels in different regions.
- Combining Semantic Guidance and Deep Reinforcement Learning For Generating Human Level Paintings29
A Semantic Guidance pipeline with a bi-level painting procedure for learning the distinction between foreground and background brush strokes at training time and introduces invariance to the position and scale of the foreground object through a neural alignment model, to zoom into a particular semantic instance.
- Effective Training Data Synthesis for Improving MLLM Chart Understanding28
This study shows that modularizing chart generation and diversifying visual details improves chart understanding capabilities, and introduces the effective chart dataset (ECD), which consistently improves the performance of various MLLMs on a range of real-world and synthetic test sets.
- Vec2Face: Scaling Face Dataset Generation with Loosely Constrained Vectors27
Vec2Face is proposed, a holistic model that uses only a sampled vector as input and can flexibly generate and control the identity of face images and their attributes, and has efficiently synthesized as many as 300K identities, whereas 60K is the largest number of identities created in the previous works.
- Confidence and Dispersity Speak: Characterising Prediction Matrix for Unsupervised Accuracy Estimation27
The nuclear norm is shown to be more accurate and robust in accuracy estimation than existing methods and the feasibility of other measurements for characterizing dispersity and confidence is validated.
- Fine-Grained Classification via Categorical Memory Networks27
This work proposes a class-specific memory module that significantly improves accuracy over baseline CNNs, achieving competitive accuracy with state-of-the-art methods on four benchmarks, including CUB-200-2011, Stanford Cars, FGVC Aircraft, and NABirds.
- Boosting Semi-Supervised Face Recognition With Noise Robustness26
A multi-agent method, named GroupNet (GN), is introduced to endow the semi-supervised face recognition solution with the ability to identify the wrongly-labelled samples and preserve the clean samples, and it is shown that GN alone achieves the leading accuracy in traditional supervised face recognition even when the noisy labels take over 50% of the training data.
- 26
- Open Set Adversarial Examples26
This work views open set recognition as a retrieval task and proposes a new approach, Opposite-Direction Feature Attack (ODFA), to generate adversarial examples / queries, and addresses the two limitations of closed set attack methods, ODFA directly works on the features for retrieval.
- Meet JEANIE: A Similarity Measure for 3D Skeleton Sequences via Temporal-Viewpoint Alignment25
JEANIE achieves state-of-the-art results on NTU-60, NTU-120, Kinetics-skeleton and UWA3D Multiview Activity II on supervised and unsupervised FSAR, and their meta-learning inspired fusion.
- SmartMask: Context Aware High-Fidelity Mask Generation for Fine-grained Object Insertion and Layout Control24
The proposed approach can also be used even without user-mask guid-ance, which allows it to perform mask-free object insertion at diverse positions and scales, and allows for better quality outputs with layout-to-image generation methods.
- Learning to Structure an Image With Few Colors24
A color quantization network, ColorCNN, is proposed, which learns to structure the images from the classification loss in an end-to-end manner, and shows superiority over other image compression methods in the extremely low bit-rate regime.
- Towards Hierarchical Multi-Agent Workflows for Zero-Shot Prompt Optimization22
This work includes a hierarchy of LLMs, first constructing a prompt with precise instructions and accurate wording in a hierarchical manner, and then using this prompt to generate the final answer to the user query, and term this pipeline Hierarchical Multi-Agent Workflow, or HMAW.
- A Locality Aware City-Scale Multi-Camera Vehicle Tracking System22
This tracking system features two parts to keep the data association within a small range, while at the same time increase the locality awareness for smaller scenarios, based on a minimal view for the large scenario.
- CIFAR-10-Warehouse: Broad and More Realistic Testbeds in Model Generalization Analysis20
This paper introduces CIFAR-10-Warehouse, consisting of 180 datasets collected by prompting image search engines and diffusion models in various ways, and aims to enhance the evaluation and deepen the understanding of two generalization tasks: domain generalization and model accuracy prediction in various out-of-distribution environments.
- 19
- Intelli-Paint: Towards Developing More Human-Intelligible Painting Agents19
This work motivates the need to learn more human-intelligible painting sequences in order to facilitate the use of autonomous painting systems in a more interactive context and proposes a novel painting approach which learns to generate output canvases while exhibiting a painting style which is more relatable to human users.
- Attribute Descent: Simulating Object-Centric Datasets on the Content Level and Beyond18
This article proposes an attribute descent approach that automatically optimizes engine attributes to enable synthetic data to approximate real-world data and confirms that adapted synthetic data can be effectively used in three scenarios: training with synthetic data only, training data augmentation and numerically understanding dataset content.
- U-Turn: Crafting Adversarial Queries with Opposite-Direction Features18
The main idea underpinning ODFA is to impel the original image feature to the opposite direction, similar to a U-turn, which shows that the adversarial queries generated by ODFA cause true matches no longer to be seen at the top ranks, and the attack success rate is consistently higher than classifier attack methods.
- Towards Optimal Feature-Shaping Methods for Out-of-Distribution Detection17
This work forms an abstract optimization framework for studying feature-shaping methods, and proposes a concrete reduction of the framework with a simple piecewise constant shaping function, which improves the generalization ability of OOD detection across a large variety of datasets and model architectures.
- 17
- 17
- 17
- Large-scale Training Data Search for Object Re-identification16
A search and pruning (SnP) solution to this training data search problem, tailored to object re-identification (re-ID), an application aiming to match the same object captured by different cameras.
- 16
- A Bag-of-Prototypes Representation for Dataset-Level Applications15
This work proposes a bag-of-prototypes (BoP) dataset representation that extends the image-level bag consisting of patch descriptors to dataset-levelbag consisting of semantic prototypes, and shows its advantage over existing representations on a series of benchmarks for two dataset- level tasks.
- 14
- 13
- 13
- 13
- Ranking Models in Unlabeled New Environments12
The problem of ranking models in unlabeled new environments is introduced, a proxy dataset that is fully labeled and well reflects the true model rankings in a given target environment is proposed, and a carefully constructed proxy set effectively captures relative performance ranking in new environments.
- 11
- 11
- Visual reranking with improved image graph11
An improved reranking method for the Bag-of-Words (BoW) based image search, based on a directed image graph robust to outlier distraction, which shows that the rank-level feature fusion can be adopted in this re ranking method as well.
- What Does Softmax Probability Tell Us about Classifiers Ranking Across Diverse Test Conditions?10
A new measure called Softmax Correlation (SoftmaxCorr) is introduced that calculates the cosine similarity between a class-class correlation matrix, constructed from Softmax output vectors across an unlabeled test dataset, and a predefined reference matrix that embodies ideal class correlations.
- 10
- 10
- Learning Camera Movement Control from Real-World Drone Videos9
DGFormer, an auto-regressive transformer that leverages the camera path and images from all past frames to predict camera movement in the next frame is introduced, which effectively learns to perform challenging camera movements such as navigating through obstacles, maintaining low altitude to increase perceived speed, and orbiting towers and buildings.
- CircNet: Meshing 3D Point Clouds with Circumcenter Detection9
This paper leverage the duality between a triangle and its circumcenter, and introduce a deep neural network that detects the circumcenters to achieve point cloud triangulation, and introduces multiple anchor priors to divide the neighborhood space of each point.
- Adaptive Affinity for Associations in Multi-Target Multi-Camera Tracking9
This paper designs experiments to verify such misfit between global re-ID feature distances and local matching in tracking, and proposes a simple yet effective approach to adapt affinity estimations to corresponding matching scopes in MTMCT.
- Spotting Visual Keywords from Temporal Sliding Windows9
This paper details the Visual Keyword Spotting system used in the first Mandarin Audio-Visual Speech Recognition Challenge (MAVSR 2019) and proposes a simple and scalable classification based strategy that achieves 19.0% mean average precision (mAP) on this challenge.
- 8
- 8
- 8
- 8
- Can We Predict Performance of Large Models across Vision-Language Tasks?7
This study proposes a new framework for predicting unknown performance scores based on observed ones from other LVLMs or tasks, and introduces several improvements to enhance probabilistic matrix factorization with Markov chain Monte Carlo.
- Optimizing Calibration by Gaining Aware of Prediction Correctness7
This paper proposes a new post-hoc calibration objective derived from the aim of calibration that asks that the calibrator decrease model confidence on wrongly predicted samples and increase confidence on correctly predicted samples.
- Bounding Box Stability against Feature Dropout Reflects Detector Generalization across Environments7
Under feature map dropout, good detectors tend to output bounding boxes whose locations do not change much, while bounding boxes of poor detectors will undergo noticeable position changes, contributing to finding that BoS score has a strong, positive correlation with detection accuracy measured by mean average precision under various test environments.
- 7
- 7
- How to Synthesize a Large-Scale and Trainable Micro-Expression Dataset?7
A protocol to automatically synthesize large scale MiE training data that allow us to train improved recognition models for real-world test data and discover three types of Action Units (AUs) that can constitute trainable MiEs.
- LeapAlign: Post-Training Flow Matching Models at Any Generation Step by Building Two-Step Trajectories6
LeapAlign is introduced, a fine-tuning method that reduces computational cost and enables direct gradient propagation from reward to early generation steps and consistently outperforms state-of-the-art GRPO-based and direct-gradient methods across various metrics, achieving superior image quality and image-text alignment.
- DiSA: Diffusion Step Annealing in Autoregressive Image Generation6
DiSA is introduced, a training-free method which gradually uses fewer diffusion steps as more tokens are generated, e.g., using 50 steps at the beginning and gradually decreasing to 5 steps at later stages, and is complementary to existing acceleration methods designed for diffusion alone.
- Training with Product Digital Twins for AutoRetail Checkout6
A training data optimization framework, i.e., training with digital twins (DtTrain), which leverage the product 3D models and optimize their rendering viewpoint and illumination to generate "digital twins" that visually resemble representative user images that better reflects the characteristics of the target scenario.
- 6
- 5
- A Study of Using Synthetic Data for Effective Association Knowledge Learning5
3D synthetic data engines can well simulate motion factors such as camera movement, camera view, and object movement so that the simulated videos can provide association modules with effective motion features, and it is shown that association knowledge obtained from synthetic data can achieve very similar performance on real-world test sets without domain adaption techniques.
- Optimizing Camera Configurations for Multi-View Pedestrian Detection5
Across multiple simulation scenarios, the configurations generated by the transformer-based model consistently outperform random search, heuristic-based methods, and configurations designed by human experts, shedding light on future camera layout optimization.
- Are Labels Necessary for Classifier Accuracy Evaluation5
This work constructs a meta-dataset: a dataset comprised of datasets generated from the original training set via various image transformations such as rotation, background substitution, foreground scaling, etc, and aims to estimate the model accuracy on unlabeled test datasets.
- Tensor index for large scale image retrieval5
A tensor index structure to improve precision and recall simultaneously is proposed and it is shown that on the one hand, the multi-index enhances the discriminative power of visual words, thus improving precision; on the otherhand, the introduction of multiple codebooks corrects quantization artifacts, so improving recall.
- DiffusionBench: On Holistic Evaluation of Diffusion Transformers4
This work introduces NanoGen, a unified DiT training and evaluation framework that supports RAE, VAE, pixel-space, and MeanFlow diffusion methods under both ImageNet and T2I setups, and summarizes ImageNet and text-to-image results, which yields DiffusionBench, a holistic benchmark for DiT research.
- MORE'24 Multimedia Object Re-ID: Advancements, Challenges, and Opportunities4
A spectrum of topics related to object re-id are covered, including but not limited to deep metric learning, multi-view data generation, video-based object re-id, cross-domain object re-id and real-world applications.
- 4
- Unsupervised Evaluation of Out-of-distribution Detection: A Data-centric Perspective4
This paper is the first to introduce the unsupervised evaluation problem in OOD detection, which aims to evaluate Ood detection methods in real-world changing environments without OOD labels, and proposes three methods to compute Gscore as an un supervised indicator of OOD Detection performance.
- Learning to Select Camera Views: Efficient Multiview Understanding at Few Glances4
This work proposes a view selection approach that analyzes the target object or scenario from given views and selects the next best view for processing and reveals that certain cameras can be shut off with minimal performance impact, shedding light on future camera layout optimization for multiview systems.
- 4
- 3
- 3
- 3
- 3
- 3
- 3
- 3
- 3
- 2
- Alice Benchmarks: Connecting Real World Re-Identification with the Synthetic2
The settings of Alice benchmarks are detailed, an analysis of existing commonly-used domain adaptation methods are provided, and an online server will be set up for the community to evaluate methods conveniently and fairly.
- Adaptive Calibrator Ensemble for Model Calibration under Distribution Shift2
This paper proposes a simple and effective method named adaptive calibrator ensemble (ACE) to calibrate OOD datasets whose difficulty is usually higher than the calibration set, and uses an adaptive weighting method that strikes a balance between the two extreme functions.
- 2
- 2
- 2
- 2
- 2
- 2
- 2
- Scene Character Recognition Using Coupled Spatial Learning2
The novel method named coupled spatial learning (CSL) for scene character representation is proposed, which simultaneously discover the spatial context in both the dictionary learning and coding stages and introduces the spatial coding strategy which utilizes the spatiality regularization to consider the relationship among features in the Euclidean space.
- Towards a Cytometry Foundation Model: Interpretable Sample-level Predictive Modelling via Pretrained Transformers1
The Generalised Pretrained Cytometry Transformer is presented, an interpretable framework designed to learn from heterogeneous marker panels for sample-level predictive modelling and significantly boosts performance on data-scarce downstream tasks, marking a pivotal step towards a cytometry foundation model.
- Gecko: A Simulation Environment with Stateful Feedback for Refining Agent Tool Calls1
Gecko, a stateful simulation environment that provides informative feedback for refining LLM tool calls before real execution, is introduced, combining rules and LLMs to check the validity of tool names and arguments, synthesize schema-conforming and state-consistent responses, and judge task completion against the user objective.
- Confidence and Dispersity as Signals: Unsupervised Model Evaluation and Ranking1
A unified and practical framework for unsupervised model evaluation and ranking in two common deployment settings: estimating the accuracy of a fixed model on multiple unlabeled test sets (dataset-centric evaluation), and ranking a set of candidate models on a single unlabeled test set (model-centric evaluation).
- 1
- Learning to Structure an Image with Few Colors and Beyond1
This work proposes a color quantization network, ColorCNN, which learns to structure an image in limited color spaces by minimizing the classification loss, and introduces ColorCNN+, which supports multiple color space size configurations, and addresses the previous issues of poor recognition accuracy and undesirable visual fidelity under large color spaces.
- 1
- 1
- –
- Combining 3D-multispectral and hyperspectral imaging to identify environmental stress treatments imposed during plant growth–
Results demonstrated the effectiveness of combining 3D-multispectral and 2D-HSI data with ML analyses for non-destructive, high-throughput phenotyping, and enabled non-destructive, high-throughput identification of environmental stress treatments imposed during plant growth.
- InpaintHuman: Reconstructing Occluded Humans with Multi-Scale UV Mapping and Identity-Preserving Diffusion Inpainting–
InpaintHuman introduces two key innovations: a multi-scale UV-parameterized representation with hierarchical coarse-to-fine feature interpolation, enabling robust reconstruction of occluded regions while preserving geometric details, and an identity-preserving diffusion inpainting module that integrates textual inversion with semantic-conditioned guidance for subject-specific, temporally coherent completion.
- –
- Gromov Wasserstein Optimal Transport for Semantic Correspondences–
This work replaces the standard nearest neighbours matching with an optimal transport algorithm that includes a Gromov Wasserstein spatial smoothness prior and shows that it can significantly boost the performance of the DINOv2 baseline, and be competitive and sometimes surpassing state-of-the-art methods using Stable Diffusion features, while being 5--10x more efficient.
- WARDEN: Endangered Indigenous Language Transcription and Translation with 6 Hours of Training Data–
This paper introduces WARDEN, an early language model system capable of transcribing and translating Wardaman, an endangered Australian indigenous language into English, and designs WARDEN to have separate transcription and translation models, which outperforms larger open-source and proprietary models and establishes a strong baseline.
- –
- –
- –
- –
- MORE'25 Multimedia Object Re-ID: Advancements, Challenges, and Opportunities–
A spectrum of topics related to object re-id are covered, including but not limited to deep metric learning, multi-view data generation, video-based object re-id, cross-domain object re-id, and real-world applications.
- –
- Flexible Geometric Guidance for Probabilistic Human Pose Estimation with Diffusion Models–
This work proposes a framework for pose estimation using diffusion models, which enables sampling from a probability distribution over plausible poses which are consistent with a 2D image, and guides samples from an unconditional diffusion model, trained only on 3D data, using the gradients of the heatmaps from a 2D keypoint detector.
- –
- –
- Assessing Model Generalization in Vicinity–
This paper proposes incorporating responses from neighboring test samples into the correctness assessment of each individual sample, and shows that applying the VRP method to existing generalization indicators, such as average confidence and effective invariance, consistently improves over these baselines both methodologically and experimentally.
- –
- –
- –
- –
- –
- Training Data Optimization in Human-Centric Analysis–
This talk focuses on a few human-centric applications: person re-identification, multi-object tracking, and micro-expression recognition, and introduces a protocol to synthesize large amounts of micro- expression training images leading to a state-of-the-art recognition system.
- –
- Multi-View Correlation Consistency for Semi-Supervised Semantic Segmentation–
This paper proposes multi-view correlation consistency (MVCC) learning: it considers rich pairwise relationships in self-correlation matrices and matches them across views to provide robust supervision and proposes a view-coherent data augmentation strategy that guarantees pixel-pixel correspondence between different views.
- Introduction to the Special Section on Learning Representations, Similarity, and Associations in Dynamic Multimedia Environments–
The aim of this Special Section on “Learning Representations, Similarities, and Associations in Dynamic Multimedia Environments” is to bring academic researchers and industry developers together for sharing the recent advances and future trends of the representation/similarity learning and association of complex multimedia data.
- Guest Editorial: Special Issue on Machine Learning and Signal Processing–
A novel unsupervised network termed GC-Net for Gaussian Curvature Optimization on images, which is composed of multiple residual convolution blocks and is trained with the loss function, which includes an image similarity term and a Gaussian curvature regularization term.
- –
- –
- –
- On the Large-Scale Transferability of Convolutional Neural Networks–
This paper investigates the effective transfer of the CNN descriptors in generic and fine-grained classification at a large scale and provides some simple yet effective methods in constructing a competitive baseline recognition system.
- Real-time Detecting System of High Knee Movement Based on Kinect–
In this study, the researchers designed and developed a real-time detecting system that can track the skeleton information of objects, and then perform action recognition and counting and it is envisioned that this real- time detecting system of high knee movement will open new possibilities for the way of counting in sports testing.
- Research on Constructing Situational Applications Based on Web2.0–
An architecture of situational application based on Web 2.0 is put forward and details on how to create an instance are given.
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 2026-10-15. 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.