Research map / Architectures and efficiency / CNN design
Belief Propagation Reloaded: Learning BP-Layers for Labeling Problems
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 BP-Layer to ESPNet (ablation) | Semantic segmentation · Cityscapes validation | mIOU: 61.4 -> 62.8 | Helped |
| Use pixel-wise pairwise scores (ablation) | LBPSS · Cityscapes validation | mIOU: 62.8 -> 63.6 | Helped |
| Jointly train ESPNet and BP-Layer (ablation) | LBPSS with pixel-wise scores · Cityscapes validation | mIOU: 63.6 -> 65.2 | Helped |
| Add multi-scale processing to BP (ablation) | Stereo BP model · Scene Flow validation | bad1: 12.6 -> 10.0 | Helped |
| Replace NLL with Huber loss (ablation) | Stereo BP+MS · Scene Flow validation | bad1: 10.0 -> 8.15 | Helped |
| Add refinement block (ablation) | Stereo BP+MS (H) · Scene Flow validation | MAE: 1.13 -> 0.74 | Helped |
| Add BP-Layer and multi-scale processing to WTA (ablation) | Optical flow model · Sintel validation, selected scenes | EPE: 1.25 -> 0.83 | Helped |
| Replace cross-entropy with Huber loss (ablation) | Optical flow BP+MS · Sintel validation, selected scenes | EPE: 0.83 -> 0.66 | Helped |
| Add refinement block (ablation) | Optical flow BP+MS (H) · Sintel validation, selected scenes | EPE: 0.66 -> 0.63 | Helped |
| Use BP+MS (H) instead of BP+MS+Ref (H) (ablation) | Stereo benchmark model · Middlebury 2014 and Kitti 2015 | bad2 / bad3 | Mixed |
Closest papers by meaning
- Belief Propagation Neural Networks (NeurIPS 2020 · similarity 0.86)
- Deep Attentive Belief Propagation: Integrating Reasoning and Learning for Solving Constraint Optimization Problems (NeurIPS 2022 · similarity 0.86)
- Learning in Deep Factor Graphs with Gaussian Belief Propagation (ICML 2024 · similarity 0.84)
- PICNN: A Pathway towards Interpretable Convolutional Neural Networks (AAAI 2024 · similarity 0.83)
- U-Nets as Belief Propagation: Efficient Classification, Denoising, and Diffusion in Generative Hierarchical Models (ICLR 2025 · similarity 0.83)
- Show and Tell: Visually Explainable Deep Neural Nets via Spatially-Aware Concept Bottleneck Models (CVPR 2025 · 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.