Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileWorks69 from public data
- MOTR: End-to-End Multiple-Object Tracking with Transformer906
MOTR is proposed, which extends DETR and introduces track query to model the tracked instances in the entire video to enhance temporal relation modeling and serve as a stronger baseline for future research on temporal modeling and Transformer-based trackers.
- PETRv2: A Unified Framework for 3D Perception from Multi-Camera Images579
PETRv2, a unified framework for 3D perception from multi-view images, is proposed and the 3D position embedding in PETR is extended for temporal modeling, which achieves the temporal alignment on object position of different frames.
- Exploring Object-Centric Temporal Modeling for Efficient Multi-View 3D Object Detection482
StreamPETR is the first online multi-view method that achieves comparable performance with lidar-based methods and is the first online multi-view method that achieves comparable performance with lidar-based methods on the standard nuScenes benchmark.
- MOTRv2: Bootstrapping End-to-End Multi-Object Tracking by Pretrained Object Detectors335
This paper first adopt the anchor formulation of queries and then use an extra object detector to generate proposals as anchors, providing detection prior to MOTR, a simple yet effective pipeline to bootstrap end-to-end multi-object tracking with a pretrained object detector.
- Learning Human-Object Interaction Detection Using Interaction Points256
This paper proposes a novel fully-convolutional approach that directly detects the interactions between human-object pairs and predicts interaction points, which directly localize and classify the inter-action.
- Panacea: Panoramic and Controllable Video Generation for Autonomous Driving209
This work proposes Panacea, an innovative approach to generate panoramic and controllable videos in driving scenarios, capable of yielding an unlimited numbers of diverse, annotated samples pivotal for autonomous driving advancements, by effectively augmenting the training dataset used for advanced BEV perception techniques.
- Cross Modal Transformer: Towards Fast and Robust 3D Object Detection193
A robust 3D detector, named Cross Modal Transformer (CMT), for end-to-end 3D multi-modal detection that achieves 74.1% NDS on nuScenes test set while maintaining faster inference speed.
- Referring Multi-Object Tracking156
This paper proposes a new and general referring understanding task, termed referring multi-object tracking (RMOT), whose core idea is to employ a language expression as a semantic cue to guide the prediction of multi- object tracking.
- Deep Contextual Attention for Human-Object Interaction Detection138
This work proposes a contextual attention framework for human-object interaction detection that leverages context by learning contextually-aware appearance features for human and object instances and adaptively selects relevant instance-centric context information to highlight image regions likely to contain human- object interactions.
- ADriver-I: A General World Model for Autonomous Driving127
This paper constructs a general world model based on MLLM and diffusion model for autonomous driving, termed ADriver-I, which takes the vision-action pairs as inputs and autoregressively predicts the control signal of the current frame.
- UniScene: Unified Occupancy-centric Driving Scene Generation125
UniScene is introduced, the first unified framework for generating three key data forms — semantic occupancy, video, and LiDAR — in driving scenes in driving scenes, and outperforms previous SOTAs in the occupancy, video, and LiDAR generation.
- 115
- 114
- Tree Energy Loss: Towards Sparsely Annotated Semantic Segmentation109
A novel tree energy loss for SASS is proposed by providing semantic guidance for unlabeled pixels by sequentially applying these affinities to the net-work prediction, achieving dynamic online self-training.
- OnlineRefer: A Simple Online Baseline for Referring Video Object Segmentation100
This work breaks up the previous offline belief and proposes a simple yet effective online model using explicit query propagation, named OnlineRefer, that leverages target cues that gather semantic information and position prior to improve the accuracy and ease of referring predictions for the current frame.
- 88
- RoboGSim: A Real2Sim2Real Robotic Gaussian Splatting Simulator82
The experimental results show that the RoboGSim data model can achieve zero-shot performance on the real robot, with results comparable to real robot data, and the real2sim and sim2real transfer experiments show a high consistency in the texture and physics.
- Forest R-CNN: Large-Vocabulary Long-Tailed Object Detection and Instance Segmentation80
Prior knowledge of the relations among object categories is exploited to cluster fine-grained classes into coarser parent classes, and construct a classification tree that is responsible for parsing an object instance into a fine- grained category via its parent class.
- GeoVLA: Empowering 3D Representations in Vision-Language-Action Models76
GeoVLA is a novel VLA framework that effectively integrates 3D information to advance robotic manipulation and achieves state-of-the-art results in the LIBERO and ManiSkill2 simulation benchmarks and shows remarkable robustness in real-world tasks requiring height adaptability, scale awareness and viewpoint invariance.
- 76
- MOTRv3: Release-Fetch Supervision for End-to-End Multi-Object Tracking75
This work proposes MOTRv3, which balances the label assignment process using the developed release-fetch supervision strategy, and achieves impressive performance across diverse benchmarks, e.g., MOT17, DanceTrack.
- Co-mining: Self-Supervised Learning for Sparsely Annotated Object Detection63
This paper proposes a simple but effective mechanism, called Co-mining, for sparsely annotated object detection, where two branches of a siamese network predict the pseudo-label sets for each other in order to enhance multi-view learning and better mine unlabeled instances.
- Reconstructive Visual Instruction Tuning60
Recon reconstructive visual instruction tuning (ROSS), a family of Large Multimodal Models (LMMs) that exploit vision-centric supervision signals, delivers competitive performance with a single SigLIP visual encoder, demonstrating the efficacy of the vision-centric supervision tailored for visual outputs.
- TopoMLP: A Simple yet Strong Pipeline for Driving Topology Reasoning58
This work introduces a powerful 3D lane detector and an improved 2D traffic element detector to extend the upper limit of topology performance and proposes TopoMLP, a simple yet high-performance pipeline for driving topology reasoning.
- Neural networks for power management optimal strategy in hybrid microgrid58
A new method by radial basis function neural network is proposed to predict the power values of renewable energy and load demand, which are used as the input values in the optimal process.
- 54
- Collective neurodynamic optimization for economic emission dispatch problem considering valve point effect in microgrid42
The collective neurodynamic optimization (CNO) method, which combines heuristic approach and projection neural network (PNN), is attempted to optimize scheduling of an electrical microgrid with ten thermal generators and minimize the plus of generation and emission cost.
- RoboChallenge: Large-scale Real-robot Evaluation of Embodied Policies37
The methodology for constructing RoboChallenge, an online evaluation system to test robotic control algorithms, and the survey of recent state-of-the-art VLA models using the authors' initial benchmark Table30 are described.
- Open-sourced Data Ecosystem in Autonomous Driving: the Present and Future35
This comprehensive review, conducted in collaboration with esteemed colleagues from both academia and industry, systematically assesses over seventy open-source autonomous driving datasets from domestic and international sources and offers insights into various aspects, such as the principles underlying the creation of high-quality datasets.
- RoboMatrix: A Skill-centric Hierarchical Framework for Scalable Robot Task Planning and Execution in Open-World33
The introduction of the first unified vision-language-action (VLA) model capable of seamlessly integrating both movement and manipulation within one model, and achieves a 50% higher success rate than task-centric baselines when applied to unseen objects, scenes, and tasks.
- LLaDA-VLA: Vision Language Diffusion Action Models32
This work presents LLaDA-VLA, the first Vision-Language-Diffusion-Action model built upon pretrained d-VLMs for robotic manipulation, and introduces two key designs: a localized special-token classification strategy that replaces full-vocabulary classification with special action token classification, reducing adaptation difficulty.
- 32
- 29
- Is a 3D-Tokenized LLM the Key to Reliable Autonomous Driving?26
This work introduces DETR-style 3D perceptrons as 3D tokenizers, which connect LLM with a one-layer linear projector and demonstrates superior performance in both 3D detection and ego planning tasks on nuScenes dataset, proving that 3D-tokenized LLM is the key to reliable autonomous driving.
- Continual LLaVA: Continual Instruction Tuning in Large Vision-Language Models24
This work proposes a new benchmark for COntinuAl inStruction Tuning on LVLMs (COAST), which encompasses the aforementioned domain-incremental, capability-incremental, and dataset-incremental configurations and proposes Continual LLaVA, a rehearsal-free method tailored for continual instruction tuning in LVLMs.
- RAGNet: Large-Scale Reasoning-Based Affordance Segmentation Benchmark Towards General Grasping22
A large-scale grasping-oriented affordance segmentation benchmark with human-like instructions, named RAGNet, and a comprehensive affordance-based grasping framework, named AffordanceNet, which consists of a VLM pretrained on the authors' massive affordance data and a grasping network that conditions an affordance map to grasp the target.
- Multi-GraspLLM: A Multimodal LLM for Multi-Hand Semantic Guided Grasp Generation20
Multi-Grasp set is presented, the first large-scale multi-hand grasp dataset with automatically contact annotations, and Multi-GraspLLM, a unified language-guided grasp generation framework, which leverages large language models (LLM) to handle variable-length sequences, generating grasp poses for diverse robotic hands in a single unified architecture.
- 19
- Glad: A Streaming Scene Generator for Autonomous Driving18
A simple yet effective framework, named Glad, to generate video data in a frame-by-frame style, which introduces a latent variable propagation module, which views the latent features of previous frame as noise prior and injects it into the latent features of current frame.
- BFA: Best-Feature-Aware Fusion for Multi-View Fine-Grained Manipulation17
This paper proposes a plug-and-play Best-Feature-Aware (BFA) fusion strategy for multi-view manipulation tasks, which is adaptable to various policies and designs a lightweight subnetwork to effectively predict the importance score of each view.
- The 1st-place Solution for CVPR 2023 OpenLane Topology in Autonomous Driving Challenge16
This work develops a multi-stage framework for high performance OpenLane Topology reasoning based on centerline detection and traffic element detection, and designs a simple yet effective MLP-based head for topology prediction.
- Complex dynamical behavior of neural networks in circuit implementation15
Several circuit units are combined to realize some neuron-network models and it is shown that the phenomenon such as hyperchaos, limit cycles, homoclinic orbits of the designed circuits are closely similar to the results of numerical experiments.
- DM0: An Embodied-Native Vision-Language-Action Model towards Physical AI13
DM0, an Embodied-Native Vision-Language-Action (VLA) framework designed for Physical AI, unifies embodied manipulation and navigation by learning from heterogeneous data sources from the onset.
- MemoryVLA++: Temporal Modeling via Memory and Imagination in Vision-Language-Action Models11
MemoryVLA++ is proposed, a full temporal modeling framework that equips VLA models with memory and imagination for robotic manipulation, and achieves strong performance across Libero, SimplerEnv, Mikasa-Robo, Calvin, Libero-Plus, and diverse real-robot tasks, validating the effectiveness of full temporal modeling with memory and imagination.
- VLM-Eval: A General Evaluation on Video Large Language Models10
This paper introduces a unified evaluation that encompasses multiple video tasks, including captioning, question and answering, retrieval, and action recognition, and proposes a simple baseline: Video-LLaVA, which uses a single linear projection and outperforms existing video LLMs.
- PriorVLA: Prior-Preserving Adaptation for Vision-Language-Action Models9
PriorVLA is proposed, a novel framework that preserves pretrained priors and learns to leverage them for effective adaptation, and keeps a frozen Prior Expert as a read-only prior source and trains an Adaptation Expert for downstream specialization.
- PADriver: Towards Personalized Autonomous Driving9
Experimental results show that PADriver outperforms state-of-the-art approaches on different evaluation metrics, and enables various driving modes.
- Neural network optimization for energy-optimal cooperative computing in wireless communication system9
A one-layer projection neural network subject to linear equalities and bound constraints is introduced and applied in mobile wireless sensor network to make it energy-efficient by reasonably allocating local and remote data sizes when processing an application within a certain period of time.
- Realtime-VLA FLASH: Speculative Inference Framework for Diffusion-based VLAs8
Realtime-VLA FLASH is proposed, a speculative inference framework that eliminates most full inference calls during replanning by introducing a lightweight draft model with parallel verification via the main model's Action Expert and a phase-aware fallback mechanism that reverts to the full inference pipeline when needed.
- 7
- SegGrasp: Zero-Shot Task-Oriented Grasping via Semantic and Geometric Guided Segmentation7
A training-free framework that incorporates both semantic and geometric priors for zero-shot task-oriented grasp generation, SegGrasp, which first leverages the vision-language models like GLIP for coarse segmentation and uses detailed geometric information from convex decomposition to improve segmentation quality through a fusion policy named GeoFusion.
- Implicit Feature Refinement for Instance Segmentation7
An empirical comparison of different refinement strategies reveals that the widely-used four consecutive convolutions are not necessary, and the implicit feature refinement (IFR) is developed by constructing an implicit function that simulates an infinite-depth refinement network.
- ManiAgent: An Agentic Framework for General Robotic Manipulation6
ManiAgent is introduced, an agentic architecture for general manipulation tasks that achieves end-to-end output from task descriptions and environmental inputs to robotic manipulation actions that yields VLA models with performance comparable to those trained on human-annotated datasets.
- 6
- ROSA: Harnessing Robot States for Vision-Language and Action Alignment6
A novel training paradigm, ROSA, which leverages robot state estimation to improve alignment between vision-language and action spaces and enables the VLA model to gain enhanced spatial understanding and self-awareness, thereby boosting performance and generalization.
- Bootstrap Masked Visual Modeling via Hard Patches Mining6
This work proposes Hard Patches Mining (HPM), predicting patch-wise losses and subsequently determining where to mask, and introduces an auxiliary loss predictor, which is trained with a relative objective to prevent overfitting to exact loss values.
- Towards 3D Object Detection with 2D Supervision6
A hybrid training framework is introduced, enabling us to learn a visual 3D object detector with massive 2D (pseudo) labels, even without 3D annotations, to break through the information bottleneck of 2D clues.
- IntentionVLA: Generalizable and Efficient Embodied Intention Reasoning for Human-Robot Interaction5
Experimental results show that IntentionVLA substantially outperforms π0, achieving 18% higher success rates with direct instructions and 28% higher than ECoT under intention instructions, and highlights IntentionVLA as a promising paradigm for next-generation human-robot interaction (HRI) systems.
- Genetic Optimization of Energy Consumption of Pellet Shaft Furnace Combustor Based on Support Vector Machine (SVM)3
The results show that the accuracy of the SVM classifier model is satisfied and the gas consumption can be reduced by 4% per ton of green pellets with this optimization method.
- DEGround: An Effective Baseline for Ego-centric 3D Visual Grounding with a Homogeneous Framework1
DEGround is proposed, which shares DETR queries as object representation for both DE tection and Ground and enables the grounding to benefit from basic category classification and box detection and introduces a regional activation grounding module that highlights instruction-related regions and a query-wise modulation module that incorporates sentence-level semantic into the query representation, strengthening the context-aware understanding of language instructions.
- 1
- –
- –
- QuadVerse: An Integrated Framework Aligning Visual-Physical Reality for Quadruped Simulation–
This paper introduces QuadVerse, an integrated framework that uses reconstructed scenes as a calibration substrate for aligning visual perception, physical interaction, and actuator dynamics, and demonstrates robust zero-shot visual-navigation policy deployment without task-specific real-world rollouts.
- –
- –
- The 1st-place Solution for ECCV 2022 Multiple People Tracking in Group Dance Challenge–
This paper presents the 1st place solution to the Group Dance Multiple People Tracking Challenge, surpassing the second-place solution by +6.8% HOTA and using the YOLOX detection proposals for the anchor initialization of detect queries.
- –
- Circuit implementation of digitally programmable transconductance amplifier in analog simulation of reaction-diffusion neural model–
The digital programmable transconductance amplifier (DPTA) is designed to realize the digitally function in circuit and applied to realize two activation functions and the simulation results are verified in Multisim.
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.