Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileWorks50 from public data
- Learning Lightweight Lane Detection CNNs by Self Attention Distillation780
It is observed that attention maps extracted from a model trained to a reasonable level would encode rich contextual information that can be used as a form of ‘free’ supervision for further representation learning through performing top- down and layer-wise attention distillation within the net- work itself.
- CLIP2Scene: Towards Label-efficient 3D Scene Understanding by CLIP273
A Semantic-driven Cross-modal Contrastive Learning framework that pre-trains a 3D network via semantic and spatial-temporal consistency regularization, which significantly outperforms other self-supervised methods and demonstrates the generalizability for handling cross-domain datasets.
- 262
- Point-to-Voxel Knowledge Distillation for LiDAR Semantic Segmentation256
This article addresses the problem of distilling knowledge from a large teacher model to a slim student network for LiDAR semantic segmentation by proposing the Point-to-Voxel Knowledge Distillation (PVD), which transfers the hidden knowledge from both point level and voxel level.
- Vision-Centric BEV Perception: A Survey245
This paper provides in-depth analyses and comparative results on various BEV perception tasks, facilitating the evaluation of future works and sparking new research directions, and discusses and shares valuable empirical implementation details to aid in the advancement of related algorithms.
- Rethinking Range View Representation for LiDAR Segmentation220
A range view method is able to surpass the point, voxel, and multi-view fusion counterparts in the competing LiDAR semantic and panoptic segmentation benchmarks, i.e., SemanticKITTI, nuScenes, and ScribbleKITTI.
- LoGoNet: Towards Accurate 3D Object Detection with Local-to-Global Cross- Modal Fusion202
The novel Local-to-Global fusion network (LoGoNet), which performs LiDAR-camerafusion at both local and global levels, and is ranked 1st on Waymo 3D object detection leaderboard and obtains 81.02 mAPH (L2) detection performance.
- Inter-Region Affinity Distillation for Road Marking Segmentation147
A novel knowledge distillation (KD) approach that can transfer ‘knowledge’ on scene structure more effectively from a teacher to a student model and consistently brings higher performance gains on all lightweight models, compared to previous distillation methods.
- SCPNet: Semantic Scene Completion on Point Cloud135
A novel knowledge distillation objective is designed, dubbed Dense-to-Sparse Knowledge Distillation (DSKD), which transfers the dense, relation-based semantic knowledge from the multi-frame teacher to the single-frame student, significantly improving the representation learning of thesingle-frame model.
- Predicting gradient is better: Exploring self-supervised learning for SAR ATR with a joint-embedding predictive architecture115
This study investigates an effective SSL method for SAR ATR, which can pave the way for a foundation model in SAR ATR and presents a novel Joint-Embedding Predictive Architecture for SAR ATR (SAR-JEPA), which leverages local masked patches to predict the multi-scale SAR gradient representations of unseen context.
- UniSeg: A Unified Multi-Modal LiDAR Segmentation Network and the OpenPCSeg Codebase113
A unified multi-modal LiDAR segmentation network, termed UniSeg, is presented, which leverages the information of RGB images and three views of the point cloud, and accomplishes semantic segmentation and panoptic segmentation simultaneously.
- SARATR-X: Toward Building a Foundation Model for SAR Target Recognition109
SARATR-X learns generalizable representations via self-supervised learning (SSL) and provides a cornerstone for label-efficient model adaptation to generic SAR target detection and classification tasks, and is trained on 0.18 M unlabelled SAR target samples, which are curated by combining contemporary benchmarks and constitute the largest publicly available dataset till now.
- Point Cloud Pre-Training with Diffusion Models82
This paper considers the point cloud pre-training task as a conditional point-to-point gen-eration problem and introduces a conditional point genera-tor that aggregates the features extracted by the backbone and employs them as the condition to guide the point-to-point recovery from the noisy point cloud, thereby assisting the backbone in capturing both local and global geometric priors as well as the global point density distri-bution of the object.
- Learning to Steer by Mimicking Features from Heterogeneous Auxiliary Networks71
This paper considerably improves the accuracy and robustness of predictions through heterogeneous auxiliary networks feature mimicking, a new and effective training method that provides us with much richer contextual signals apart from steering direction.
- STCrowd: A Multimodal Dataset for Pedestrian Perception in Crowded Scenes67
A novel method, Density-aware Hierarchical heatmap Aggregation (DHA), to enhance pedestrian perception in crowded scenes is proposed and achieves state-of-the-art performance for pedestrian detection on various datasets.
- 54
- A Comprehensive Survey on 3D Content Generation50
A new taxonomy is proposed that categorizes existing approaches into three types: 3D native generative methods, 2D prior-based 3D generative methods, and hybrid 3D generative methods.
- Categorical Relation-Preserving Contrastive Knowledge Distillation for Medical Image Classification49
A novel Categorical Relation-preserving Contrastive Knowledge Distillation (CRCKD) algorithm, which takes the commonly used mean-teacher model as the supervisor, is proposed, which can distill the relational knowledge more comprehensively.
- OccMamba: Semantic Occupancy Prediction with State Space Models42
This work presents the first Mamba-based network for semantic occupancy prediction, termed OccMamba, inspired by the global modeling and linear computation complexity of the Mamba architecture, and presents a simple yet effective 3D-to-1D reordering scheme, i.e., height-prioritized 2D Hilbert expansion.
- 38
- Human-centric Scene Understanding for 3D Large-scale Scenarios38
This paper presents a large-scale multi-modal dataset for human-centric scene under-standing, dubbed HuCenLife, which is collected in diverse daily-life scenarios with rich and fine-grained annotations and designs novel modules for LiDAR-based segmentation and action recognition.
- 38
- Frozen CLIP Transformer Is an Efficient Point Cloud Encoder33
This paper introduces Efficient Point Cloud Learning (EPCL), an effective and efficient point cloud learner for directly training high-quality point cloud models with a frozen CLIP transformer and achieves promising performance on both indoor and outdoor benchmarks.
- See More and Know More: Zero-shot Point Cloud Segmentation via Multi-modal Visual Data32
This work proposes a novel multi-modal zero-shot learning method to better utilize the complementary information of point clouds and images for more accurate visual-semantic alignment.
- Network pruning via resource reallocation32
PEEL first constructs a predefined backbone and then conducts resource reallocation on it to shift parameters from less informative layers to more important layers in one round, thus amplifying the positive effect of these informative layers.
- Uni3D-LLM: Unifying Point Cloud Perception, Generation and Editing with Large Language Models29
The efficacy of Uni3D-LLM in the comprehension, generation, and editing of point cloud has been validated and the impact of integrating a point cloud perception module on the generation and editing processes is assessed, confirming the substantial potential of the approach for practical applications.
- TASeg: Temporal Aggregation Network for LiDAR Semantic Segmentation28
A Temporal LiDAR Aggregation and Distillation (TLAD) algorithm, which leverages historical priors to assign dif-ferent aggregation steps for different classes, and a Static-Moving Switch Augmentation (SMSA) algorithm, which utilizes sufficient temporal information to enable objects to switch their motion states freely, thus greatly increasing static and moving training samples.
- 26
- 19
- Gradient modulated contrastive distillation of low-rank multi-modal knowledge for disease diagnosis17
This work proposes a novel Momentum-enriched Multi-Modal Low-Rank (M3LR) constraint to explore the high-order correlations and complementary information among different modalities, thus yielding more accurate multi-modal diagnosis.
- 14
- Semi-supervised 3D Object Detection with PatchTeacher and PillarMix12
This paper proposes PatchTeacher, a semi-supervised 3D object detection method that focuses on partial scene 3D object detection to provide high-quality pseudo labels for the student, and introduces three key techniques, i.e., Patch Normalizer, Quadrant Align, and Fovea Selection, to improve the performance of PatchTeacher.
- CluB: Cluster Meets BEV for LiDAR-Based 3D Object Detection12
This paper proposes a new 3D object detection framework, referred to as CluB, which incorporates an auxiliary cluster-based branch into the BEV-based detector by enriching the object representation at both feature and query levels.
- 12
- RangePerception: Taming LiDAR Range View for Efficient and Accurate 3D Object Detection11
Two novel algorithms named Range Aware Kernel (RAK) and Vision Restoration Module (VRM), which facilitate information from range image representation and world-coordinate 3D detection results are proposed, which facilitate information extraction from range image representation and world-coordinate 3D detection results.
- Advances in 3D pre-training and downstream tasks: a survey9
This paper provides a comprehensive survey of recent advances in the 3D feld, which encompasses a wide collection of topics, including diverse pre-training strategies, backbone designs and downstream tasks.
- 8
- MoE3D: Mixture of Experts meets Multi-Modal 3D Understanding7
This paper proposes MoE3D, which integrates Mixture of Experts (MoE) into the multi-modal learning framework, and deploys a set of specialized "expert" networks, each adept at processing a specific modality or a mode of cross-modal interaction.
- EvoDefense: Co-Evolving Black-Box Defense with Large Language Models3
EvoDefense is introduced, an experience-guided co-evolving black-box defense paradigm that employs a guard LLM to detect malicious queries and an experience memory module to accumulate defense knowledge from previous interactions that enables EvoDefense to generalize across unseen attacks and target models without retraining.
- Segment and Select: Vision-Language Segmentation in 3D Scenarios2
This paper proposes the SEGment-And-select (SEGA3D) paradigm for 3D visionlanguage segmentation that directly operates on the fine-grained visual information and is free from the superpoint dependency.
- Learning Multi-Modal Trajectory Policies for Data-Efficient Robotic Manipulation2
This work proposes MATE (Multi-ModAl TrajEctory Policies), a novel trajectory prediction framework built upon Mixture-of-Experts (MoE), and introduces a Multi-Modal MoE architecture to achieve fine-grained sub-token feature decoupling, and designs a cross-modal cosine router for stable and scale-invariant expert assignment across heterogeneous modalities.
- 2
- CMTFormer: Marrying Transformer with Hierarchical Information Interaction for RGB-Event Object Detection–
This paper proposes a Cross-Modal information inTeraction transFormer, coined as CMTFormer, which hierarchically integrates RGB and event information to achieve efficient and stable multimodal collaboration and designs a shallow-to-deep information interaction scheme.
- Heterogeneous and Adept Snapshot Distillation for 3D Semantic Segmentation–
Information-oriented Heterogeneous Distillation (IHD) is presented to help the uni-modal model absorb the complementary knowledge from the multi-modal teacher and the Information-Oriented Filtering (IOF) strategy to select informative images from the continuous image sequence for multi-modal fusion is designed.
- –
- –
- –
- PVCap: Towards Accurate 3D Dense Captioning via PseudoCap and VoxelCapNet–
The proposed PVCap consists of PseudoCap and VoxelCapNet, a robust caption network that utilizes voxel features and adapts the caption head to the voxel-based network architecture and can serve as a competitive baseline for future research on 3D dense captioning.
- –
- –
Publication data from OpenAlex; 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.