Research map / 3D vision / Depth and stereo
Light Field Intrinsics With a Deep Encoder-Decoder Network
Read the paperCheck 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 |
|---|---|---|---|
| Use the proposed encoder-decoder instead of ACC | light-field disparity estimation · Specular synthetic scenes | MSE (depth) ×100: 5.9 vs 23.1; 4.6 vs 13.4 | Helped |
| Use the proposed encoder-decoder instead of EPI1 or EPI2 | light-field disparity estimation · Specular synthetic scenes | MSE (depth) ×100: 5.9 vs 30.0 and 35.5; 4.6 vs 14.7 and 17.3 | Helped |
| Use the proposed method instead of Alperovich [2] or Sulc et al. [34] | light-field intrinsic decomposition · Nine unseen test datasets | GMSE ×100 specular: 0.23 vs 1.04; 0.23 vs 1.01 | Helped |
| Use the proposed method instead of Shi et al. [31] | light-field intrinsic decomposition · Nine unseen test datasets | LMSE ×100 diffuse: 0.15 vs 0.34 | Hurt |
| Use the proposed method instead of Shi et al. [31] | light-field intrinsic decomposition · Nine unseen test datasets | LMSE ×100 specular: 0.11 vs 0.15 | Hurt |
| Use the proposed method instead of Alperovich [2], Sulc et al. [34], or Shi et al. [31] | light-field intrinsic decomposition · Nine unseen test datasets | SSIM ×100 diffuse: 80.08 vs 74.98, 75.43, and 63.02 | Helped |
| Use the proposed method instead of Alperovich [2], Sulc et al. [34], or Shi et al. [31] | light-field intrinsic decomposition · Nine unseen test datasets | SSIM ×100 specular: 81.37 vs 48.46, 47.25, and 73.71 | Helped |
| Reduce spatial patch size to 24 × 24 (ablation) | proposed light-field network · Nine unseen test datasets | LMSE ×100 diffuse: 0.25 -> 0.33 | Hurt |
| Reduce spatial patch size to 24 × 24 (ablation) | proposed light-field network · Nine unseen test datasets | LMSE ×100 specular: 0.19 -> 0.33 | Hurt |
| Reduce spatial patch size to 24 × 24 (ablation) | proposed light-field network · Specular synthetic scenes | MSE (depth) ×100: 5.9 -> 192.7; 4.6 -> 167.9 | Hurt |
| Train only on the center view, without disparity information from sub-aperture views (ablation) | proposed light-field network · Specular synthetic scenes | MSE (depth) ×100: 5.9 -> 55.87; 4.6 -> 19.31 | Hurt |
Closest papers by meaning
- EPINET: A Fully-Convolutional Neural Network Using Epipolar Geometry for Depth From Light Field Images (CVPR 2018 · similarity 0.90)
- Light Field Reconstruction Using Deep Convolutional Network on EPI (CVPR 2017 · similarity 0.89)
- What Sparse Light Field Coding Reveals About Scene Structure (CVPR 2016 · similarity 0.89)
- Memory-oriented Decoder for Light Field Salient Object Detection (NeurIPS 2019 · similarity 0.88)
- NeLF-Pro: Neural Light Field Probes for Multi-Scale Novel View Synthesis (CVPR 2024 · similarity 0.85)
- TurboSL: Dense Accurate and Fast 3D by Neural Inverse Structured Light (CVPR 2024 · similarity 0.85)
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.