Research map / 3D vision / Depth and stereo
Deep Material Recognition in Light-Fields via Disentanglement of Spatial and Angular Information
Bichuan Guo, Jiangtao Wen, Yuxing Han
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 sparse directional sub-views (model-4) (ablation) | ang-filter · LFMR test | accuracy (%): 77.83 -> 78.91 | Improved |
| add angular rotational invariance (weight sharing and directional pooling) (ablation) | model-4 · LFMR test | accuracy (%): 78.91 -> 80.74 | Improved |
| add angular registration (ablation) | model-4 · LFMR test | accuracy (%): 78.91 -> 79.36 | Improved |
| add representative selection (ablation) | model-4 · LFMR test | accuracy (%): 78.91 -> 79.85 | Improved |
| add both registration and selection (ablation) | model-4 · LFMR test | accuracy (%): 78.91 -> 80.38 | Improved |
| add both registration and selection (ablation) | model-S · LFMR test | accuracy (%): 80.74 -> 81.75 | Improved |
| select region randomly instead of by depth criterion (ablation) | model-S · LFMR test | accuracy (%): 81.75 -> 81.41 | Hurt |
| warp pixels with estimated depth instead of selecting a region (ablation) | model-S · LFMR test | accuracy (%): 81.45 -> 80.35 | Hurt |
| use 1x1 angular kernel instead of 3x3 (ablation) | angular-S · LFMR validation | accuracy (%): 81.16 -> 81.92 | Improved |
| increase angular filters from 16 to 32 to 64 (ablation) | angular-S · LFMR validation | accuracy (%): 81.35 -> 81.92 -> 80.93 | Mixed |
| use representative region grid N=3 instead of whole patch (ablation) | angular-S with spatial · LFMR validation | accuracy (%): 81.45 -> 81.92 | Improved |
| crop angular component into smaller regions (ablation) | angular features only · LFMR validation | accuracy (%): 66.34 -> 51.93 | Hurt |
Closest papers by meaning
- A 4D Light-Field Dataset and CNN Architectures for Material Recognition (ECCV 2016 · similarity 0.90)
- Differential Angular Imaging for Material Recognition (CVPR 2017 · similarity 0.89)
- Differential Viewpoints for Ground Terrain Material Recognition (TPAMI 2020 · similarity 0.87)
- Learning Light Field Angular Super-Resolution via a Geometry-Aware Network (AAAI 2020 · similarity 0.86)
- Deep Spatial-angular Regularization for Compressive Light Field Reconstruction over Coded Apertures (ECCV 2020 · similarity 0.85)
- Neural LightRig: Unlocking Accurate Object Normal and Material Estimation with Multi-Light Diffusion (CVPR 2025 · similarity 0.84)
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.
