Research map / Point clouds / Segmentation and detection
Retro-FPN: Retrospective Feature Pyramid Network for Point Cloud Semantic Segmentation
Peng Xiang, Xin Wen, Yu-Shen Liu, Hui Zhang, Yi Fang, Zhizhong Han
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 hierarchical supervision alone without retro-transformer (ablation) | Point Transformer · S3DIS Area 5 | mIoU: 70.4 -> 70.6 | No clear effect |
| add retro-transformer (cross-att, posemb, gate) without hierarchical supervision (ablation) | Point Transformer · S3DIS Area 5 | mIoU: 70.4 -> 70.8 | No clear effect |
| add local cross-attention with hierarchical supervision (ablation) | Point Transformer · S3DIS Area 5 | mIoU: 70.6 -> 71.9 | Improved |
| add learnable position embedding (ablation) | Retro-FPN cross-attention · S3DIS Area 5 | mIoU: 71.9 -> 72.4 | Improved |
| add semantic gate unit (ablation) | Retro-FPN cross-attention · S3DIS Area 5 | mIoU: 71.9 -> 72.2 | Improved |
| combine position embedding and semantic gate (ablation) | Retro-FPN · S3DIS Area 5 | mIoU: 71.9 -> 73.0 | Improved |
| plug Retro-FPN into KPConv deform | KPConv deform · S3DIS Area 5 | mIoU: 67.1 -> 70.7 | Improved |
| plug Retro-FPN into MinkowskiNet | MinkowskiNet · SemanticKITTI test | mIoU: 64.1 -> 68.0 | Improved |
| plug Retro-FPN into Point Transformer V2 | Point Transformer V2 · ScanNet v2 val | mIoU: 75.4 -> 76.0 | Improved |
| plug Retro-FPN into Point Transformer | Point Transformer · S3DIS 6-fold | mIoU: 73.5 -> 77.3 | Improved |
| add Retro-FPN to Point Transformer | Point Transformer · S3DIS Area 5 | inference latency (s): 54.05 -> 55.16 | Hurt |
Closest papers by meaning
- 3D Recurrent Neural Networks with Context Fusion for Point Cloud Semantic Segmentation (ECCV 2018 · similarity 0.87)
- SQN: Weakly-Supervised Semantic Segmentation of Large-Scale 3D Point Clouds (ECCV 2022 · similarity 0.87)
- EPSegFZ: Efficient Point Cloud Semantic Segmentation for Few- and Zero-Shot Scenarios with Language Guidance (AAAI 2026 · similarity 0.87)
- Associatively Segmenting Instances and Semantics in Point Clouds (CVPR 2019 · similarity 0.87)
- GrowSP: Unsupervised Semantic Segmentation of 3D Point Clouds (CVPR 2023 · similarity 0.87)
- Semantic Segmentation for Real Point Cloud Scenes via Bilateral Augmentation and Adaptive Fusion (CVPR 2021 · similarity 0.87)
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.
