Research map / Detection and segmentation / Weakly supervised detection and segmentation
What Synthesis Is Missing: Depth Adaptation Integrated With Weak Supervision for Indoor Scene Parsing
Keng-Chi Liu, Yi-Ting Shen, Jan P. Klopp, Liang-Gee Chen
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 min-max normalization to the depth adversarial loss (ablation) | depth adaptation · SUN RGB-D val (raw depth) | mIoU (w/o windows): 26.44 -> 30.95 | Improved |
| add sensor noise simulation to synthetic depth (inpainted) (ablation) | depth-only scene parsing · SUN RGB-D val (inpainted depth) | mIoU (w/o windows): 28.42 -> 34.86 | Improved |
| add sensor noise simulation to synthetic depth (raw) (ablation) | depth-only scene parsing · SUN RGB-D val (raw depth) | mIoU (w/o windows): 22.22 -> 30.95 | Improved |
| use proposed noise simulation instead of prior simulation (ablation) | depth-only scene parsing · SUN RGB-D val (inpainted depth) | mIoU (w/o windows): 32.61 -> 34.86 | Improved |
| add SceneNet to Pbrs synthetic training data (ablation) | depth-adapted teacher · SUN RGB-D val | mIoU: 29.24 -> 32.25 | Improved |
| add RGB student with first-stage depth-contour pseudo-labels (ablation) | teacher · SUN RGB-D val | mIoU: 32.25 -> 35.21 | Improved |
| add second integration stage with weak localization (ablation) | first-stage pseudo-labels · SUN RGB-D val | mIoU: 35.21 -> 43.53 | Improved |
| add first-stage depth cue to weak localization (ablation) | second-stage-only labels · SUN RGB-D val | mIoU: 25.97 -> 43.53 | Improved |
| add second stage on bed class (ablation) | pseudo-label integration · SUN RGB-D val | bed IoU: 54.93 -> 52.06 | Hurt |
| train student network on pseudo-labels instead of using pseudo-labels directly | scene parsing student · SUN RGB-D train | effective mIoU: 41.64 -> 49.46 | Improved |
| add UCM contour refinement at inference (ablation) | Ours (Full) · SUN RGB-D val | mIoU: 43.53 -> 43.80 | Improved |
| use weak image-level labels with depth adaptation instead of CYCADA | scene parsing · SUN RGB-D val | mIoU: 28.90 -> 43.53 | Improved |
Closest papers by meaning
- Indoor Scene Parsing With Instance Segmentation, Semantic Labeling and Support Relationship Inference (CVPR 2017 · similarity 0.88)
- Understanding Real World Indoor Scenes With Synthetic Data (CVPR 2016 · similarity 0.88)
- Hypersim: A Photorealistic Synthetic Dataset for Holistic Indoor Scene Understanding (ICCV 2021 · similarity 0.87)
- 3D-to-2D Distillation for Indoor Scene Parsing (CVPR 2021 · similarity 0.87)
- SceneNet RGB-D: Can 5M Synthetic Images Beat Generic ImageNet Pre-Training on Indoor Segmentation? (ICCV 2017 · similarity 0.87)
- InstructScene: Instruction-Driven 3D Indoor Scene Synthesis with Semantic Graph Prior (ICLR 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.
