Research map / 3D vision / Depth and stereo
Differential Viewpoints for Ground Terrain Material Recognition
Jia Xue, Hang Zhang, Ko Nishino, Kristin Dana
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 differential image stream (single view DAIN) instead of single view CNN | VGG-M CNN · GTOS | accuracy (%): 74.3 -> 79.4 | Improved |
| use differential image Idelta instead of second raw view Iv+delta as second input (ablation) | single view DAIN · GTOS | accuracy (%): 77.5 -> 79.4 | Improved |
| use Sum instead of Max for feature-map combination (ablation) | single view DAIN · GTOS | accuracy (%): 79.4 -> 79.0 | No clear effect |
| use hybrid intermediate-and-final layer combination instead of final only (ablation) | two-stream DAIN · GTOS | accuracy (%): 77.0 -> 79.4 | Improved |
| use hybrid combination instead of intermediate only (ablation) | two-stream DAIN · GTOS | accuracy (%): 74.8 -> 79.4 | Improved |
| add multiview input (4 views) to single view DAIN | DAIN VGG-M · GTOS | accuracy (%): 79.4 -> 81.2 | Improved |
| use pooling instead of voting for multiview DAIN (ablation) | DAIN MobileNet V2 · GTOS | accuracy (%): 85.8 -> 86.2 | Improved |
| use 3D filter instead of Sum for multiview pooling (ablation) | multiview DAIN · GTOS | accuracy (%): 81.2 -> 81.1 | No clear effect |
| replace VGG-M backbone with MobileNet V2 | DAIN · GTOS | accuracy (%): 81.2 -> 86.2 | Improved |
| add DEP texture encoding and pooling to MobileNet V2 | MobileNet V2 · GTOS | accuracy (%): 80.4 -> 83.3 | Improved |
| add TEAN (texture encoded angular network) to DAIN | multiview DAIN · GTOS | accuracy (%): 86.2 -> 87.6 | Improved |
| add multi-scale training to TEAN (ablation) | TEAN · GTOS (split 1) | accuracy (%): 82.87 -> 81.93 | Hurt |
Closest papers by meaning
- Differential Angular Imaging for Material Recognition (CVPR 2017 · similarity 0.97)
- Deep Texture Manifold for Ground Terrain Recognition (CVPR 2018 · similarity 0.87)
- Deep Material Recognition in Light-Fields via Disentanglement of Spatial and Angular Information (ECCV 2020 · similarity 0.87)
- Geometry-Informed Material Recognition (CVPR 2016 · similarity 0.84)
- Infusing Synthetic Data with Real-World Patterns for Zero-Shot Material State Segmentation (NeurIPS 2024 · similarity 0.83)
- Patched Line Segment Learning for Vector Road Mapping (AAAI 2024 · similarity 0.82)
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.
