Research map / 3D vision / 3D generation
Efficient Semantic Scene Completion Network with Spatial Group Convolution
Jiahui Zhang, Hao Zhao, Anbang Yao, Yurong Chen, Li Zhang, Hongen Liao
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 |
|---|---|---|---|
| use the proposed sparse conv network with multiscale encoder-decoder instead of SSCNet | SGC sparse conv network · SUNCG | semantic scene completion avg IoU: 46.4 -> 70.5 | Improved |
| add SGC with 4 groups (fixed pattern partition) to baseline | SGC sparse conv network · SUNCG | semantic scene completion avg IoU: 70.5 -> 69.3 with FLOPs 79G -> 22G | Hurt |
| add SGC with 2 random groups | SGC sparse conv network · SUNCG | semantic scene completion avg IoU: 70.5 -> 69.9 with FLOPs 79G -> 42G | Hurt |
| add SGC with 6 groups (pattern) | SGC sparse conv network · SUNCG | semantic scene completion avg IoU: 70.5 -> 66.9 with FLOPs 79G -> 16G | Hurt |
| use fixed pattern partition instead of random partition with 4 groups (ablation) | SGC sparse conv network · SUNCG | semantic scene completion avg IoU: 67.6 -> 69.3 | Improved |
| use fixed pattern partition instead of random partition with 3 groups (ablation) | SGC sparse conv network · SUNCG | scene completion IoU: 82.6 -> 84.1 | Improved |
| use fixed pattern partition instead of random with 2 groups (ablation) | SGC sparse conv network · SUNCG | semantic scene completion avg IoU: 69.9 -> 69.6 | Mixed |
| add SGC with 4 groups (pattern) effect on small categories like chair | SGC sparse conv network · SUNCG | per-class IoU drop: -3.7 chair, -2.1 objs vs 0.2 ceil | Hurt |
| use fixed pattern partition instead of random, G=2 (ablation) | SGC sparse conv network · NYU | semantic scene completion IoU: 24.1 -> 26.5 | Improved |
| use fixed pattern partition instead of random, G=4 (ablation) | SGC sparse conv network · NYU | semantic scene completion IoU: 22.6 -> 25.9 | Improved |
| add SGC G=3 pattern partition to baseline | SGC sparse conv network · NYU | semantic scene completion IoU: 26.4 -> 26.7 | No clear effect |
| add random-partition SGC G=3 to baseline | SGC sparse conv network · NYU | semantic scene completion IoU: 26.4 -> 23 | Hurt |
Closest papers by meaning
- Semantic Scene Completion From a Single Depth Image (CVPR 2017 · similarity 0.90)
- Anisotropic Convolutional Networks for 3D Semantic Scene Completion (CVPR 2020 · similarity 0.88)
- RGBD Based Dimensional Decomposition Residual Network for 3D Semantic Scene Completion (CVPR 2019 · similarity 0.88)
- Cascaded Context Pyramid for Full-Resolution 3D Semantic Scene Completion (ICCV 2019 · similarity 0.88)
- Anisotropic Convolutional Neural Networks for RGB-D based Semantic Scene Completion (TPAMI 2021 · similarity 0.88)
- Context and Geometry Aware Voxel Transformer for Semantic Scene Completion (NeurIPS 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.
