Research map / 3D vision / Depth and stereo
Epipolar Consistent Attention Aggregation Network for Unsupervised Light Field Disparity Estimation
Chen Gao, Shuo Zhang, Youfang Lin
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 |
|---|---|---|---|
| replace linear upsampling with deconvolution (epipolar consistent upsample) (ablation) | ECAAN5 · SLFD Lion | MSE x100 / Bp(0.3): 2.48/14.69 -> 2.06/6.79 | Improved |
| integrate at attention level with linear upsampling instead of averaging disparities (ablation) | ECAAN5 · SLFD Lion | MSE x100 / Bp(0.3): 4.96/21.26 -> 2.48/14.69 | Improved |
| add COFA occlusion-free aggregation block (ablation) | ECAAN5 · SLFD Lion | MSE x100 / Bp(0.3): 2.06/6.79 -> 0.08/1.13 | Improved |
| add occlusion masks to the unsupervised loss (versus none) (ablation) | ECAAN5 · SLFD Lion | MSE x100 / Bp(0.3): 0.26/2.17 -> 0.08/1.13 | Improved |
| use full occlusion masks instead of only sub-occlusion masks (ablation) | ECAAN5 · SLFD Lion | MSE x100 / Bp(0.3): 0.16/1.46 -> 0.08/1.13 | Improved |
| use all 5x5 views instead of 3x3 views | ECAAN · SLFD Lion | MSE x100: 0.173 -> 0.087 | Improved |
| use all 5x5 views instead of 3x3 views | ECAAN · SLFD Toy bricks | MSE x100: 0.332 -> 0.408 | Hurt |
| use ECAAN3 instead of UnMVFM | ECAAN3 · SLFD Rooster clock | MSE x100: 0.261 -> 0.177 | Improved |
Closest papers by meaning
- Attention-Based View Selection Networks for Light-Field Disparity Estimation (AAAI 2020 · similarity 0.89)
- Sheared Epipolar Focus Spectrum for Dense Light Field Reconstruction (TPAMI 2023 · similarity 0.88)
- Fast Light-Field Disparity Estimation With Multi-Disparity-Scale Cost Aggregation (ICCV 2021 · similarity 0.88)
- Heterogeneous Light Fields (CVPR 2016 · similarity 0.87)
- Attention-based Multi-Level Fusion Network for Light Field Depth Estimation (AAAI 2021 · similarity 0.87)
- From Pixels to Views: Learning Angular-Aware and Physics-Consistent Representations for Light Field Microscopy (NeurIPS 2025 · 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.
