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
- Antonio Manuel LópezSuggested from co-authorship
Is this you? Claim this profile to confirm or dismiss it.
Works185 from public data
- ERFNet: Efficient Residual Factorized ConvNet for Real-Time Semantic Segmentation1,484
A deep architecture that is able to run in real time while providing accurate semantic segmentation, and a novel layer that uses residual connections and factorized convolutions in order to remain efficient while retaining remarkable accuracy is proposed.
- 928
- Dreaming to Distill: Data-Free Knowledge Transfer via DeepInversion738
DeepInversion is introduced, a new method for synthesizing images from the image distribution used to train a deep neural network, which optimizes the input while regularizing the distribution of intermediate feature maps using information stored in the batch normalization layers of the teacher.
- See through Gradients: Image Batch Recovery via GradInversion648
It is shown that gradients encode a surprisingly large amount of information, such that all the individual images can be recovered with high fidelity via GradInversion, even for complex datasets, deep networks, and large batch sizes.
- A-ViT: Adaptive Tokens for Efficient Vision Transformer606
This work introduces A - ViT, a method that adaptively adjusts the inference cost of vision transformer (ViT) for images of different complexity by automatically reducing the number of tokens in vision transformers that are processed in the network as inference proceeds, and demonstrates that it requires no extra parameters or sub-network for halting.
- 513
- VoxFormer: Sparse Voxel Transformer for Camera-Based 3D Semantic Scene Completion479
VoxFormer, a Transformer-based semantic scene completion framework that can output complete 3D volumetric semantics from only 2D images that outperforms the state of the art with a relative improvement of 20.0% in geometry and 18.1% in semantics.
- Cost Volume Pyramid Based Depth Inference for Multi-View Stereo429
It is demonstrated that building a cost volume pyramid in a coarse-to-fine manner instead of constructing a costvolume at a fixed resolution leads to a compact, lightweight network and allows us inferring high resolution depth maps to achieve better reconstruction results.
- Less Is More: Towards Compact CNNs343
This work shows that, by incorporating sparse constraints into the objective function, it is possible to decimate the number of neurons during the training stage, thus theNumber of parameters and the memory footprint of the neural network are reduced, which is desirable at the test time.
- 336
- Is Ego Status All You Need for Open-Loop End-to-End Autonomous Driving?296
This paper conducts thorough analyses and introduces a new metric to evaluate whether the predicted trajectories adhere to the road, and proposes a simple baseline able to achieve competitive results without relying on perception annotations.
- Hydra-MDP: End-to-end Multimodal Planning with Multi-target Hydra-Distillation288
Hydra-MDP, a novel paradigm employing multiple teachers in a teacher-student model, which features a multi-head decoder to learn diverse trajectory candidates tailored to various evaluation metrics, achieves the 1-st place in the Navsim challenge.
- Understanding The Robustness in Vision Transformers268
This paper examines the role of self-attention in learning robust representations in Vision Transformers and proposes a family of fully attentional networks (FANs) that strengthen this capability by incorporating an attentional channel processing design.
- Context Based Emotion Recognition using EMOTIC Dataset266
This paper presents EMOTIC, a dataset of images of people in a diverse set of natural situations, annotated with their apparent emotion, and trains different CNN models for emotion recognition, combining the information of the bounding box containing the person with the contextual information extracted from the scene.
- Road Scene Segmentation from a Single Image261
It is concluded that convolutional neural networks are suitable for learning 3D scene layout from noisy labels and provides a relative improvement of 7% compared to the baseline and combining color planes provides a statistical description of road areas that exhibits maximal uniformity.
- M$^2$BEV: Multi-Camera Joint 3D Detection and Segmentation with Unified Birds-Eye View Representation249
A unified framework that jointly performs 3D object detection and map segmentation in the Birds Eye View~(BEV) space with multi-camera image inputs, which is memory efficient, allowing significantly higher resolution images as input, with faster inference speed.
- FB-OCC: 3D Occupancy Prediction based on Forward-Backward View Transformation222
The proposed solution FB-OCC builds upon FB-BEV, a cutting-edge camera-based bird's-eye view perception design using forward-backward projection, and results in a state-of-the-art mIoU score of 54.19% on the nuScenes dataset.
- Learning Image Matching by Simply Watching Video217
An unsupervised learning based approach to the ubiquitous computer vision problem of image matching that achieves surprising performance comparable to traditional empirically designed methods.
- 215
- OmniDrive: A Holistic Vision-Language Dataset for Autonomous Driving with Counterfactual Reasoning212
This work proposes OmniDrive, a holistic vision-language dataset that aligns agent models with 3D driving tasks through counter-factual reasoning, and explores two advanced OmniDrive-Agent frameworks to assess the importance of vision-language alignment versus 3D perception.
- Panoptic SegFormer: Delving Deeper into Panoptic Segmentation with Transformers199
This work presents Panoptic SegFormer, a general framework for panoptic segmentation with transformers, which contains three innovative components: an efficient deeply-supervised mask decoder, a query decoupling strategy, and an improved postprocessing method.
- FB-BEV: BEV Representation from Forward-Backward View Transformations171
This work proposes a novel forward-backward view transformation module that compensates for the deficiencies in both existing methods, allowing them to enhance each other to obtain higher quality BEV representations mutually.
- Effective Use of Synthetic Data for Urban Scene Semantic Segmentation164
A drastically different way to handle synthetic images that does not require seeing any real images at training time is introduced, which builds on the observation that foreground and background classes are not affected in the same manner by the domain shift, and thus should be treated differently.
- 164
- FasterViT: Fast Vision Transformers with Hierarchical Attention158
The newly introduced Hierarchical Attention (HAT) approach decomposes global self-attention with quadratic complexity into a multi-level attention with reduced computational costs and can be used as a plug-and-play module for existing networks and enhance them.
- Active Learning for Deep Object Detection via Probabilistic Modeling158
The method uses a scoring function that aggregates these two types of uncertainties for both heads to obtain every image’s informativeness score and outperforms single-model based methods and performs on par with multi- model based methods at a fraction of the computing cost.
- FreeSOLO: Learning to Segment Objects without Annotations155
This work proposes a fully unsupervised learning method that learns class-agnostic instance segmentation without any annotations, and presents a novel localization-aware pre-training framework, where objects can be discovered from complicated scenes in an unsuper supervised manner.
- Built-in Foreground/Background Prior for Weakly-Supervised Semantic Segmentation154
This work proposes a novel method to extract markedly more accurate masks from the pre-trained network itself, forgoing external objectness modules, and introduces a new form of inexpensive weak supervision yielding an additional accuracy boost.
- FocalFormer3D : Focusing on Hard Instance for 3D Object Detection133
This work proposes Hard Instance Probing (HIP), a general pipeline that identifies FN in a multi-stage manner and guides the models to focus on excavating difficult instances, and instantiates this method as FocalFormer3D, a simple yet effective detector that excels at excavating difficult objects and improving prediction recall.
- Scalable Active Learning for Object Detection128
This paper has built a scalable production system for active learning in the domain of autonomous driving, and describes the resulting high-level design, sketch some of the challenges and their solutions, and present the current results at scale.
- 125
- 124
- 120
- Bridging the Day and Night Domain Gap for Semantic Segmentation110
Diverse options such as enlarging the dataset to cover these domains in unsupervised training or adapting the images on-the-fly during inference to a comfortable domain such as sunny daylight in a pre-processing step are explored, allowing IV perception systems to work reliably also at night.
- 106
- 90
- 89
- 89
- Class Weighted Convolutional Features for Visual Instance Search81
This paper proposes a local-aware encoding of convolutional features based on semantic information predicted in the target image and obtains the most discriminative regions of an image using Class Activation Maps (CAMs), which is based on the knowledge contained in the network and has the additional advantage of not requiring external information.
- BEVNeXt: Reviving Dense BEV Frameworks for 3D Object Detection80
This paper aims to address the drawbacks of the existing dense BEV-based 3D object detectors by introducing the proposed enhanced components, including a CRF-modulated depth estimation module enforcing object-level consistencies, a long-term temporal aggregation module with extended receptive fields, and a two-stage object decoder combining perspective techniques with CRF-modulated depth embedding.
- When to Prune? A Policy towards Early Structural Pruning80
This work introduces an Early Pruning Indicator (EPI) that relies on sub-network architectural similarity and quickly triggers pruning when the sub- network's architecture stabilizes, and shows that EPI empowers a quick tracking of early training epochs suitable for pruning, offering same efficacy as an otherwise “oracle” grid-search that scans through epochs and requires orders of magnitude more compute.
- Eagle 2: Building Post-Training Data Strategies from Scratch for Frontier Vision-Language Models77
This work addresses VLM post-training from a data-centric perspective, showing the key role of data strategy in developing frontier VLMs and leads to a family of performant VLMs named Eagle2.
- Generalized Trajectory Scoring for End-to-end Multimodal Planning73
GTRS (Generalized Trajectory Scoring) is a unified framework for end-to-end multi-modal planning that combines coarse and fine-grained trajectory evaluation and demonstrates superior performance even with sub-optimal sensor inputs.
- 71
- Self-supervised Learning of Depth Inference for Multi-view Stereo66
This work proposes a self-supervised learning framework for multi-view stereo that exploit pseudo labels from the input data that outperforms existing unsupervised multi-View stereo networks by a large margin and performs on par compared to the supervised counterpart.
- Domain-Adaptive Deep Network Compression64
The Domain Adaptive Low Rank (DALR) method significantly outperforms existing low-rank compression techniques and takes into account the target domain, so it can more optimally remove the redundancy in the weights.
- 59
- Semantic Road Segmentation via Multi-scale Ensembles of Learned Features58
An algorithm based on convolutional neural networks is proposed to learn local features from training data at different scales and resolutions using a weighted linear combination and its performance is similar to state---of---the---art methods using other sources of information such as depth, motion or stereo.
- Not All Labels Are Equal: Rationalizing The Labeling Costs for Training Object Detection55
This work proposes a unified frame-work for active learning, that considers both the uncertainty and the robustness of the detector, ensuring that the network performs well in all classes, and leverages auto-labeling to suppress a potential distribution drift while boosting the performance of the model.
- Hydra-NeXt: Robust Closed-Loop Driving with Open-Loop Training51
This paper introduces Hydra-NeXt, a novel multi-branch planning framework that unifies trajectory prediction, control prediction, and a trajectory refinement network in one model and proposes the Trajectory Refinement module to augment and refine the planning decisions by effectively adhering to kinematic constraints in closed-loop environments.
- 51
- 50
- Centaur: Robust End-to-End Autonomous Driving with Test-Time Training49
Centaur (Cluster Entropy for Test-time trAining using Uncertainty) is proposed, which updates a planner's behavior via test-time training, without relying on hand-engineered rules or cost functions, and measures and minimize the uncertainty in the planner's decisions.
- Non-parametric Depth Distribution Modelling based Depth Inference for Multi-view Stereo47
This work proposes constructing the cost volume by non-parametric depth distribution modeling to handle pixels with unimodal and multi-modal distributions and introduces a sparse cost aggregation network to derive information within each volume.
- 47
- Bringing Background into the Foreground: Making All Classes Equal in Weakly-Supervised Video Semantic Segmentation41
This paper introduces an approach to weak supervision using only image tags by making use of classifier heatmaps, and develops a two-stream deep architecture that jointly leverages appearance and motion, and design a loss based on the authors' heatmaps to train it.
- 41
- Incorporating Network Built-in Priors in Weakly-Supervised Semantic Segmentation40
This work proposes a novel method to extract accurate masks from networks pre-trained for the task of object recognition, thus forgoing external objectness modules, and shows how foreground/background masks can be obtained from the activations of higher-level convolutional layers of a network.
- 39
- How Much More Data Do I Need? Estimating Requirements for Downstream Tasks39
This work considers a broad class of computer vision tasks and systematically investigates a family of functions that generalize the power-law function to allow for better estimation of data requirements, and shows that incorporating a tuned correction factor and collecting over multiple rounds significantly improves the performance of the data estimators.
- 39
- VideoITG: Multimodal Video Understanding with Instructed Temporal Grounding36
This work designs the VidThinker pipeline, which automates annotation by generating instruction-conditioned captions, retrieving relevant video segments, and selecting key frames to enable efficient supervision, and builds the VideoITG-40K dataset, which consistently boosts the performance on multiple multimodal video understanding benchmarks.
- 36
- 34
- 34
- 32
- 31
- Semantic labeling for prosthetic vision30
This work introduces semantic labeling as a technique to improve navigation outcomes for prosthetic vision users, and produces a novel egocentric vision dataset to demonstrate how semantic labeling can be applied to this problem.
- 28
- Shadow Resistant Road Segmentation from a Mobile Monocular System28
This paper proposes segmentation in a shadowless image which is computable in real---time using a color camera and argues that by using a different feature space to perform the segmentation the authors can minimize the problem of shadows from the very beginning.
- 27
- 26
- Vision Transformers are Good Mask Auto-Labelers25
Qualitative results indicate that masks produced by MAL are, in some cases, even better than human annotations, and state-of-the-art box-supervised methods by significant margins.
- 25
- Expression-Invariant Age Estimation25
This method jointly learns the age and expression by introducing a new graphical model with a latent layer between the age/expression labels and the features which captures the face changes which induce the aging and expression appearance, and thus obtaining expression-invariant age estimation.
- 24
- Knowledge Distillation for 6D Pose Estimation by Aligning Distributions of Local Predictions22
This work proposes to distill the teacher's distribution of local predictions into the student network, facilitating its training and yields state-of-the-art results with different compact student models and for both keypoint-based and dense prediction-based architectures.
- Quadtree Generating Networks: Efficient Hierarchical Scene Parsing with Sparse Convolutions22
This paper presents Quadtree Generating Networks (QGNs), a novel approach able to drastically reduce the memory footprint of modern semantic segmentation networks, using quadtrees to represent the predictions and target segmentation masks instead of dense pixel grids.
- Towards Viewpoint Robustness in Bird’s Eye View Segmentation21
This paper introduces a technique for novel view synthesis and uses it to transform collected data to the viewpoint of target rigs, allowing us to train BEV segmentation models for diverse target rigs without any additional data collection or labeling cost.
- 20
- 19
- 19
- 18
- VACL: Variance-Aware Cross-Layer Regularization for Pruning Deep Residual Networks17
This paper introduces a novel approach which puts the ith filters of layers connected by skip-connections into one regularization group and defines Variance-Aware Cross-Layer (VACL) regularization which takes into account both the first and second-order statistics of the connected layers to constrain the variance within a group.
- 17
- 16
- 16
- Learning Photometric Invariance for Object Detection15
The proposed fusion method is robust to data uncertainty and produces properly diversified color invariant ensembles, and is extended to deal with temporal data by predicting the evolution of observations over time.
- 14
- 14
- Less is More: An Exploration of Data Redundancy with Active Dataset Subsampling.14
This paper proposes an approach called Active Dataset Subsampling (ADS), to identify favorable subsets within a dataset for training using ensemble based uncertainty estimation, and finds that there are low uncertainty subsets, which can be as large as 50% of the full dataset, that negatively impact performance.
- 14
- 13
- Improving Distant 3D Object Detection Using 2D Box Supervision12
This work proposes LR3D, a frame-work that learns to recover the missing depth of distant ob-jects and adopts an implicit projection head to learn the generation of mapping between 2D boxes and depth using the 3D supervision on close objects.
- Learning Camera Pose from Optical Colonoscopy Frames Through Deep Convolutional Neural Network (CNN)12
It is shown that transfer learning from surgical simulation to actual endoscopic based surgery is a possible approach for deep learning technologies and could be estimated with higher accuracy and speed than feature based computer vision methods such as the classical structure from motion pipeline.
- Enhancing Autonomous Driving Safety with Collision Scenario Integration11
SafeFusion is proposed, a training framework to learn from collision data that integrates safety-oriented metrics during training to enable collision avoidance learning and proposes CollisionGen, a scalable data generation pipeline to generate diverse, high-quality scenarios using natural language prompts, generative models, and rule-based filtering.
- 11
- SSE: Multimodal Semantic Data Selection and Enrichment for Industrial-scale Data Assimilation10
This work quantitatively shows that the Semantic Selection and Enrichment framework (SSE) can a) successfully maintain model performance with a smaller training dataset and b) improve model performance by enriching the smaller dataset without exceeding the original dataset size.
- 10
- 9
- 9
- 9
- Fast-dVLM: Efficient Block-Diffusion VLM via Direct Conversion from Autoregressive VLM8
A suite of multimodal diffusion adaptations, block size annealing, causal context attention, auto-truncation masking, and vision efficient concatenation, that collectively enable effective block diffusion in the VLM setting are introduced.
- What is Point Supervision Worth in Video Instance Segmentation?8
This work reduces the human annotations to only one point for each object in a video frame during training, and obtains high-quality mask predictions close to fully supervised models.
- 8
- Fully Attentional Networks with Self-emerging Token Labeling7
This paper revisits the FAN models and improves their pretraining with a self-emerging token labeling (STL) framework, demonstrating significantly enhanced performance on downstream tasks such as semantic segmentation, with up to 1.7% improvement over the counterpart model.
- 7
- 7
- 7
- 7
- 6
- 6
- 6
- 6
- 5
- 5
- 5
- 5
- 5
- 5
- 5
- 5
- 5
- 4
- 4
- 4
- 4
- 3
- 3
- 3
- 3
- 3
- Mitigating Multimodal Hallucinations via Gradient-based Self-Reflection2
Gradient-based Influence-Aware Constrained Decoding (GACD), an inference-based method, that addresses both biases without auxiliary models, and is readily applicable to existing models without finetuning, is proposed.
- 2
- 2
- 2
- 2
- 2
- 2
- 2
- 1
- 1
- 1
- 1
- Exploring Camera Encoder Designs for Autonomous Driving Perception1
This work starts from a standard general-purpose encoder, ConvNeXt and progressively transforms the design, resulting in an architecture optimized for AV camera encoder achieving 8.79% mAP improvement over the baseline.
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- ExpandNets: Exploiting Linear Redundancy to Train Small Networks.–
This paper proposes to expand each linear layer of a small network into multiple linear layers, without adding any nonlinearity, so that the resulting expanded network can be compressed back to the small one algebraically, but consistently outperforms training the small network from scratch.
- –
- –
- –
- –
- –
- –
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 2026-10-16. 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.