Research map / 3D vision / 3D generation
Learning Long-Range Information with Dual-Scale Transformers for Indoor Scene Completion
Ziqi Wang, Fei Luo, Xiaoxiao Long, Wenxiao Zhang, Chunxia Xiao
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 |
|---|---|---|---|
| replace the global structure extractor with a CNN (ablation) | DST-Net stage 1 · Matterport3D | Chamfer Distance (x10^-1): 0.22 -> 0.32 | Hurt |
| replace the region geometry generator with a CNN (ablation) | DST-Net stage 1 · Matterport3D | Chamfer Distance (x10^-1): 0.22 -> 0.26 | Hurt |
| add contrastive attention training (CAT) (ablation) | DST-Net stage 1 · Matterport3D | Chamfer Distance (x10^-1): 0.24 -> 0.22 | Improved |
| train stage 2 on fused target (stage-1 output fused with target) instead of GT (ablation) | DST-Net full model · Matterport3D | Chamfer Distance (x10^-1): 0.46 -> 0.23 | Improved |
| add stage 1 before stage 2 (versus stage 2 only) (ablation) | DST-Net · Matterport3D | Chamfer Distance (x10^-1): 0.49 -> 0.23 | Improved |
| add stage 2 refinement on top of stage 1 (ablation) | DST-Net · Matterport3D | Chamfer Distance (x10^-1): 0.22 -> 0.23 (CD); recall 0.77 -> 0.78 | No clear effect |
| add CIoT loss (ablation) | DST-Net stage 1 · Matterport3D | Chamfer Distance (x10^-1): 0.25 -> 0.22 | Improved |
| add structure loss LS (ablation) | DST-Net stage 1 · Matterport3D | Chamfer Distance (x10^-1): 0.26 -> 0.22 | Improved |
| add depth loss LD to the other losses (ablation) | DST-Net stage 1 · Matterport3D | Chamfer Distance (x10^-1): 0.29 -> 0.22 | Improved |
| use DST-Net instead of SG-NN on Matterport3D precision | scene completion · Matterport3D (394 test rooms) | Precision: 0.62 -> 0.61 | Hurt |
| reduce input frames to 20% at test time (ablation) | DST-Net · Matterport3D | Chamfer Distance (x10^-1): 0.23 (40%) -> 0.47 (20%) | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- SPSG — SPSG: Self-Supervised Photometric Scene Generation From RGB-D Scans (CVPR 2021) · better on 5 of 5 shared results
- SG-NN — SG-NN: Sparse Generative Neural Networks for Self-Supervised Scene Completion of RGB-D Scans (CVPR 2020) · better on 1 of 3 shared results
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
- DeCoTR: Enhancing Depth Completion with 2D and 3D Attentions (CVPR 2024 · similarity 0.88)
- CIRCLE: Convolutional Implicit Reconstruction and Completion for Large-Scale Indoor Scene (ECCV 2022 · similarity 0.88)
- AdaSFormer: Adaptive Serialized Transformers for Monocular Semantic Scene Completion from Indoor Environments (CVPR 2026 · similarity 0.88)
- Context and Geometry Aware Voxel Transformer for Semantic Scene Completion (NeurIPS 2024 · similarity 0.87)
- H2GFormer: Horizontal-to-Global Voxel Transformer for 3D Semantic Scene Completion (AAAI 2024 · similarity 0.86)
- IPFormer: Visual 3D Panoptic Scene Completion with Context-Adaptive Instance Proposals (NeurIPS 2025 · 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.
