Research map / Point clouds / Segmentation and detection
Diffusion-Based Contextual Reconstruction for Point Cloud Segmentation with Limited Annotations
Jiawei Lian, Zhengxue Wang, Wentao Qu, Haobo Jiang, Le Hui, Jian Yang
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 the CGL module (local plus global guidance) to the diffusion framework (ablation) | DiCoSeg · S3DIS Area5 (0.1% labels) | mIoU: 64.2 -> 68.3 | Improved |
| add Local Context Aggregation to the guidance (ablation) | DiCoSeg · S3DIS Area5 (0.1% labels) | mIoU: 64.2 -> 67.4 | Improved |
| add Global Geometric Fusion to the guidance (ablation) | DiCoSeg · S3DIS Area5 (0.1% labels) | mIoU: 64.2 -> 66.9 | Improved |
| combine LCA and GGF rather than using either alone (ablation) | DiCoSeg · ScanNetV2 val (0.1% labels) | mIoU: 69.7 -> 70.7 | Improved |
| add the CGL module to the diffusion framework (ablation) | DiCoSeg · ScanNetV2 val (0.1% labels) | mIoU: 66.0 -> 70.7 | Improved |
| share only the decoder instead of sharing encoder and decoder (ablation) | DiCoSeg · S3DIS Area5 (0.1% labels) | mIoU: 64.3 -> 68.3 | Improved |
| share only the decoder instead of only the encoder (ablation) | DiCoSeg · ScanNetV2 val (0.1% labels) | mIoU: 65.6 -> 70.7 | Improved |
| add diffusion-based contextual reconstruction to SparseUNet under full supervision | SparseUNet · S3DIS Area5 (100% labels) | mIoU: 65.4 -> 68.6 | Improved |
| use DiCoSeg instead of AADNet at very low label rate | SemanticKITTI weak supervision · SemanticKITTI seq08 (0.1% labels) | mIoU: 53.3 -> 62.6 | Improved |
| use DiCoSeg instead of PointCT at 0.1% labels | ScanNetV2 weak supervision · ScanNetV2 val (0.1% labels) | mIoU: 63.7 -> 70.3 | Improved |
| use DiCoSeg instead of PointCT at 0.1% labels on S3DIS | S3DIS weak supervision · S3DIS Area5 (0.1% labels) | mIoU: 68.3 -> 68.3 | No clear effect |
Closest papers by meaning
- An End-to-End Robust Point Cloud Semantic Segmentation Network with Single-Step Conditional Diffusion Models (CVPR 2025 · similarity 0.90)
- Images as Noisy Labels: Unleashing the Potential of the Diffusion Model for Open-Vocabulary Semantic Segmentation (ICCV 2025 · similarity 0.89)
- Point Cloud Pre-training with Diffusion Models (CVPR 2024 · similarity 0.89)
- SQN: Weakly-Supervised Semantic Segmentation of Large-Scale 3D Point Clouds (ECCV 2022 · similarity 0.89)
- Point Cloud Semantic Segmentation with Sparse and Inhomogeneous Annotations (AAAI 2025 · similarity 0.89)
- SeaLion: Semantic Part-Aware Latent Point Diffusion Models for 3D Generation (CVPR 2025 · similarity 0.88)
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.
