Research map / Point clouds / Shape and surface completion
On the Over-Smoothing Problem of CNN Based Disparity Estimation
Chuangrong Chen, Xiaozhi Chen, Hui Cheng
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 full-band weighted average with single-modal weighted average at inference (ablation) | PSMNet stack-hourglass (L_reg) · Sceneflow | 3px SEE5: 9.40 -> 4.17 | Improved |
| replace full-band with single-modal weighted average at inference (ablation) | PSMNet basic (L_reg) · Sceneflow | 3px SEE5: 10.97 -> 5.36 | Improved |
| replace full-band with single-modal weighted average at inference (ablation) | GCNet (L_reg) · Sceneflow | 3px SEE5: 9.81 -> 5.39 | Improved |
| replace full-band with single-modal weighted average at inference (ablation) | PDSNet (L_reg) · Sceneflow | 3px SEE5: 12.11 -> 6.01 | Improved |
| replace full-band with single-modal weighted average at inference (ablation) | PSMNet stack-hourglass (L_reg) · Middlebury | 3px SEE5: 23.97 -> 17.94 | Improved |
| train with Gaussian cross entropy instead of smooth L1 regression loss (single-modal inference) (ablation) | PSMNet stack-hourglass · Sceneflow | 3px SEE5: 4.17 -> 2.53 | Improved |
| train with Gaussian cross entropy instead of smooth L1 regression loss (single-modal inference) (ablation) | PSMNet stack-hourglass · Sceneflow | 3px EPE: 2.83 -> 2.21 | Improved |
| use Gaussian instead of Laplace ground-truth distribution in cross entropy (ablation) | GCNet · Sceneflow | 3px SEE5: 6.91 -> 3.92 | Improved |
| use Gaussian instead of Laplace ground-truth distribution in cross entropy (ablation) | PDSNet · Sceneflow | 3px SEE5: 10.04 -> 3.43 | Improved |
| use Laplace cross entropy instead of smooth L1 loss (ablation) | GCNet (single-modal) · Sceneflow | 3px SEE5: 5.39 -> 6.91 | Hurt |
| use Laplace cross entropy instead of smooth L1 loss (ablation) | PDSNet (single-modal) · Sceneflow | 3px SEE5: 6.01 -> 10.04 | Hurt |
| lower the variance of the Gaussian ground-truth distribution (ablation) | PSMNet stack-hourglass · Sceneflow | 3px SEE5 | Improved |
Closest papers by meaning
- Adaptive Multi-Modal Cross-Entropy Loss for Stereo Matching (CVPR 2024 · similarity 0.85)
- EDNet: Efficient Disparity Estimation With Cost Volume Combination and Attention-Based Spatial Residual (CVPR 2021 · similarity 0.85)
- Learning on the Edge: Investigating Boundary Filters in CNNs (IJCV 2019 · similarity 0.85)
- Efficient Edge-Preserving Multi-View Stereo Network for Depth Estimation (AAAI 2023 · similarity 0.84)
- Pixel-Perfect Depth with Semantics-Prompted Diffusion Transformers (NeurIPS 2025 · similarity 0.84)
- DiffusionEdge: Diffusion Probabilistic Model for Crisp Edge Detection (AAAI 2024 · similarity 0.83)
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.
