Research map / Point clouds / Segmentation and detection
ProtoTransfer: Cross-Modal Prototype Transfer for Point Cloud Segmentation
Pin Tang, Hai-Ming Xu, Chao Ma
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 prototype-bank knowledge transfer from multi-modal fusion to single-modal LiDAR (ProtoTransfer vs 2DPASS) | LiDAR point cloud segmentation · nuScenes test | mIoU: 80.8 -> 82.1 | Improved |
| add prototype-bank knowledge transfer vs 2DPASS | LiDAR point cloud segmentation · SemanticKITTI test | mIoU: 72.9 -> 73.6 | Improved |
| add Lproto with a static (non-updated) prototype bank (ablation) | baseline · nuScenes val | mIoU: 76.21 -> 76.50 | No clear effect |
| dynamically update the prototype bank with fusion and unmatched-pixel features (ablation) | baseline + Lproto · nuScenes val | mIoU: 76.50 -> 80.51 | Improved |
| build prototype bank from image features only (ablation) | ProtoTransfer · nuScenes val | mIoU: 76.21 -> 75.35 | Hurt |
| build prototype bank from point features only (ablation) | ProtoTransfer · nuScenes val | mIoU: 76.21 -> 76.97 | Improved |
| build prototype bank from matched fusion features only (ablation) | ProtoTransfer · nuScenes val | mIoU: 76.21 -> 78.65 | Improved |
| naively sum point, image and fusion features into prototype bank instead of fusing class-mean embeddings via shared MLP (ablation) | ProtoTransfer · nuScenes val | mIoU: 80.51 -> 75.58 | Hurt |
| include fused unmatched-pixel class-mean embedding in prototype bank (full bank inputs) (ablation) | ProtoTransfer · nuScenes val | mIoU: 78.65 -> 80.51 | Improved |
| directly add unmatched pixel features to the prototype bank without class-mean fusion (ablation) | ProtoTransfer · nuScenes val | mIoU | No clear effect |
| set balance weight lambda to 0.2 (weight fused unmatched embedding more) (ablation) | ProtoTransfer · nuScenes val | mIoU | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- 2DPASS — 2DPASS: 2D Priors Assisted Semantic Segmentation on LiDAR Point Clouds (ECCV 2022) · better on 2 of 3 shared results
- RPVNet — RPVNet: A Deep and Efficient Range-Point-Voxel Fusion Network for LiDAR Point Cloud Segmentation (ICCV 2021) · 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
- Uni-to-Multi Modal Knowledge Distillation for Bidirectional LiDAR-Camera Semantic Segmentation (TPAMI 2024 · similarity 0.90)
- Drive&Segment: Unsupervised Semantic Segmentation of Urban Scenes via Cross-Modal Distillation (ECCV 2022 · similarity 0.89)
- LiDAR-Based Panoptic Segmentation via Dynamic Shifting Network (CVPR 2021 · similarity 0.89)
- Unsupervised Semantic Segmentation of Urban Scenes via Cross-Modal Distillation (IJCV 2025 · similarity 0.89)
- 2DPASS: 2D Priors Assisted Semantic Segmentation on LiDAR Point Clouds (ECCV 2022 · similarity 0.89)
- Delivering Arbitrary-Modal Semantic Segmentation (CVPR 2023 · similarity 0.89)
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.
