Research map / 3D vision / Depth and stereo
PS-FCN: A Flexible Learning Framework for Photometric Stereo
Guanying Chen, Kai Han, Kwan-Yee K. Wong
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 |
|---|---|---|---|
| use max-pooling instead of average-pooling for fusion (train 32, test 100) (ablation) | PS-FCN · Synthetic test | mean angular error (deg): 5.52 -> 5.30 | Improved |
| use max-pooling instead of average-pooling for fusion (train 16, test 16) (ablation) | PS-FCN · Synthetic test | mean angular error (deg): 6.70 -> 6.24 | Improved |
| use max-pooling instead of average-pooling (train 32, test 8) (ablation) | PS-FCN · Synthetic test | mean angular error (deg): 10.94 -> 11.84 | Hurt |
| use max-pooling instead of convolutional fusion (test 32) (ablation) | PS-FCN · Synthetic test | mean angular error (deg): 6.49 -> 5.50 | Improved |
| add sculpture dataset to blobby training (train 32, test 100) (ablation) | PS-FCN · Synthetic test | mean angular error (deg): 5.30 -> 4.55 | Improved |
| add sculpture dataset to blobby training (train 8, test 8) (ablation) | PS-FCN · Synthetic test | mean angular error (deg): 7.40 -> 7.20 | Improved |
| increase training inputs per sample from 8 to 32 (blobby, test 100) (ablation) | PS-FCN · Synthetic test | mean angular error (deg): 5.82 -> 5.30 | Improved |
| train with single input per sample instead of 8 or more (test 100) (ablation) | PS-FCN · Synthetic test | mean angular error (deg): 5.82 -> 13.67 | Hurt |
| increase number of test images from 8 to 100 (train 16, blobby) | PS-FCN · Synthetic test | mean angular error (deg): 8.44 -> 5.43 | Improved |
| train with 32 inputs but test with 1 image (blobby) | PS-FCN · Synthetic test | mean angular error (deg): 5.30 -> 45.17 | Hurt |
| test with 96 images instead of 16 on DiLiGenT main | PS-FCN (B+S+32) · DiLiGenT main | mean angular error (deg): 9.37 -> 8.39 | Improved |
| replace DPSN with PS-FCN (B+S+32, 96) | DiLiGenT main · DiLiGenT main | mean angular error (deg): 9.41 -> 8.39 | Improved |
Closest papers by meaning
- Deep Photometric Stereo for Non-Lambertian Surfaces (TPAMI 2020 · similarity 0.97)
- CNN-PS: CNN-based Photometric Stereo for General Non-Convex Surfaces (ECCV 2018 · similarity 0.93)
- Deep Learning Methods for Calibrated Photometric Stereo and Beyond (TPAMI 2024 · similarity 0.92)
- Self-Calibrating Deep Photometric Stereo Networks (CVPR 2019 · similarity 0.92)
- A CNN Based Approach for the Point-Light Photometric Stereo Problem (IJCV 2022 · similarity 0.91)
- Uncalibrated Neural Inverse Rendering for Photometric Stereo of General Surfaces (CVPR 2021 · similarity 0.90)
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.
