Research map / 3D vision / Depth and stereo
PX-NET: Simple and Efficient Pixel-Wise Training of Photometric Stereo Networks
Fotios Logothetis, Ignas Budvytis, Roberto Mecca, Roberto Cipolla
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 saturation variation to training data (ablation) | PX-CNN-PS · DiLiGenT (average, PX-CNN-PS ablation) | mean angular error (deg): 11.48 -> 10.97 | Improved |
| add shadow modelling (ablation) | PX-CNN-PS · DiLiGenT (average, PX-CNN-PS ablation) | mean angular error (deg): 10.97 -> 7.78 | Improved |
| add ambient light modelling (ablation) | PX-CNN-PS · DiLiGenT (average, PX-CNN-PS ablation) | mean angular error (deg): 7.78 -> 7.18 | Improved |
| add self-reflection modelling (ablation) | PX-CNN-PS · DiLiGenT (average, PX-CNN-PS ablation) | mean angular error (deg): 7.18 -> 6.83 | Improved |
| add surface discontinuity modelling (ablation) | PX-CNN-PS · DiLiGenT (average, PX-CNN-PS ablation) | mean angular error (deg): 6.83 -> 6.79 | Improved |
| add ambient and reflection effects (Ball object) (ablation) | PX-CNN-PS · DiLiGenT Ball | mean angular error (deg): 2.1 -> 2.4 | Hurt |
| use improved PX-NET architecture instead of CNN-PS architecture (ablation) | PX-CNN-PS · DiLiGenT (average, PX-CNN-PS ablation) | mean angular error (deg): 6.79 -> 6.43 | Improved |
| add RGB channels to observation maps (ablation) | PX-NET · DiLiGenT (average, PX-CNN-PS ablation) | mean angular error (deg): 6.43 -> 6.28 | Improved |
| compute self reflection with different material per reflecting point instead of same material (ablation) | PX-CNN-PS · DiLiGenT (average, PX-CNN-PS ablation) | mean angular error (deg): 6.83 -> 7.26 | Hurt |
| test-time rotation augmentation K=10 instead of K=1 | PX-NET · DiLiGenT (dense lights, average) | mean angular error (deg): 6.28 -> 6.17 | Improved |
| test-time rotation augmentation K=10 instead of K=1 | PX-NET · DiLiGenT (10 random lights, average) | mean angular error (deg): 8.76 -> 8.37 | Improved |
| train CNN-PS on more globally rendered objects (beyond 30) (ablation) | CNN-PS · DiLiGenT (average) | mean angular error (deg): 7.11 | No clear effect |
Later papers that compared with this method
1 later accepted paper reports PX-NET as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- A CNN Based Approach for the Point-Light Photometric Stereo Problem (IJCV 2022 · better on 1 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SPLINE-Net — SPLINE-Net: Sparse Photometric Stereo Through Lighting Interpolation and Normal Estimation Networks (ICCV 2019) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Universal Photometric Stereo Network Using Global Lighting Contexts (CVPR 2022 · similarity 0.90)
- CNN-PS: CNN-based Photometric Stereo for General Non-Convex Surfaces (ECCV 2018 · similarity 0.90)
- A CNN Based Approach for the Point-Light Photometric Stereo Problem (IJCV 2022 · similarity 0.90)
- Neural Inverse Rendering for General Reflectance Photometric Stereo (ICML 2018 · similarity 0.89)
- GPS-Net: Graph-based Photometric Stereo Network (NeurIPS 2020 · similarity 0.89)
- What is Learned in Deep Uncalibrated Photometric Stereo? (ECCV 2020 · similarity 0.89)
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.
