Research map / Point clouds / Point cloud classification
HVDistill: Transferring Knowledge from Images to Point Clouds via Unsupervised Hybrid-View Distillation
Sha Zhang, Jiajun Deng, Lei Bai, Houqiang Li, Wanli Ouyang, Yanyong Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add BEV distillation to IPV-only distillation (ablation) | HVDistill · nuScenes semseg fine-tune (1% labels) | mIoU: 41.5 -> 42.7 | Improved |
| add IPV distillation to BEV-only distillation (ablation) | HVDistill · nuScenes semseg fine-tune (1% labels) | mIoU: 37.8 -> 42.7 | Improved |
| pretrain with hybrid-view distillation vs random init | HVDistill · nuScenes semseg fine-tune (1% labels) | mIoU: 28.3 -> 42.7 | Improved |
| pretrain with HVDistill vs SLidR on SemanticKITTI | HVDistill · SemanticKITTI (1% labels) | mIoU: 44.6 -> 49.7 | Improved |
| supervise image-to-BEV depth with projected points (ablation) | HVDistill · nuScenes semseg fine-tune (1% labels) | mIoU: 39.6 -> 42.7 | Improved |
| supervise depth by projected points rather than only concatenating sparse depth map (ablation) | HVDistill · nuScenes semseg fine-tune (1% labels) | mIoU: 42.2 -> 42.7 | Improved |
| use superpixel-to-superpoint instead of pixel-to-point contrast on IPV (ablation) | HVDistill · nuScenes semseg fine-tune (1% labels) | mIoU: 41.1 -> 42.7 | Improved |
| add IPV contrast at all vs none (ablation) | HVDistill · nuScenes semseg fine-tune (1% labels) | mIoU: 33.4 -> 42.7 | Improved |
| increase pretraining data from 5% to 100% of nuScenes (ablation) | HVDistill · nuScenes semseg fine-tune (1% labels) | mIoU: 35.7 -> 42.7 | Improved |
| use DINO teacher instead of MoCoV2 (ablation) | HVDistill · nuScenes semseg fine-tune (1% labels) | mIoU: 42.7 -> 41.4 | Hurt |
| use supervised ImageNet teacher instead of MoCoV2 (ablation) | HVDistill · nuScenes semseg fine-tune (1% labels) | mIoU: 42.7 -> 42.9 | No clear effect |
| change alpha:beta loss weight ratio from 1:1 to 4:1 (ablation) | HVDistill · SemanticKITTI (1% labels) | mIoU: 49.4 -> 50.1 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- PVRCNN++ — PV-RCNN: Point-Voxel Feature Set Abstraction for 3D Object Detection (CVPR 2020) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Image2Point: 3D Point-Cloud Understanding with 2D Image Pretrained Models (ECCV 2022 · similarity 0.88)
- Multi-View Representation is What You Need for Point-Cloud Pre-Training (ICLR 2024 · similarity 0.88)
- Sonata: Self-Supervised Learning of Reliable Point Representations (CVPR 2025 · similarity 0.87)
- LinNet: Linear Network for Efficient Point Cloud Representation Learning (NeurIPS 2024 · similarity 0.87)
- Dataset Distillation of 3D Point Clouds via Distribution Matching (NeurIPS 2025 · similarity 0.86)
- Mitigating Object Dependencies: Improving Point Cloud Self-Supervised Learning through Object Exchange (CVPR 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
