Research map / 3D vision / Depth and stereo
Deep Coarse-to-Fine Dense Light Field Reconstruction With Flexible Sampling and Geometry-Aware Fusion
Jing Jin, Junhui Hou, Jie Chen, Huanqiang Zeng, Sam Kwong, Jingyi Yu A
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 refinement module (ablation) | Ours · HCI 4(a)->7x7 | PSNR/SSIM: 35.60/0.954 -> 36.54/0.961 | Improved |
| add refinement module (ablation) | Ours · HCI 4(f)->7x7 | PSNR/SSIM: 37.33/0.965 -> 38.68/0.971 | Improved |
| add refinement module (ablation) | Ours · 30scenes 4(a)->7x7 | PSNR/SSIM: 40.12/0.979 -> 41.18/0.982 | Improved |
| add refinement module (ablation) | Ours · HCI 3(g)->7x7 | PSNR/SSIM: 35.39/0.953 -> 36.38/0.960 | Improved |
| use confidence-based blending instead of convolutional blending (ablation) | Ours (no refinement) · 30scenes 4(a)->7x7 | PSNR/SSIM: 39.78/0.978 -> 40.12/0.979 | Improved |
| use confidence-based blending instead of convolutional blending (ablation) | Ours (no refinement) · Occlusions 4(f)->7x7 | PSNR/SSIM: 37.34/0.974 -> 38.14/0.976 | Improved |
| use confidence-based blending instead of convolutional blending (ablation) | Ours (no refinement) · Reflective 4(f)->7x7 | PSNR/SSIM: 38.59/0.955 -> 39.24/0.957 | Improved |
| use PSV-based disparity estimation instead of plain convolution on input SAIs (ablation) | Ours · HCI disparity maps | disparity error | Improved |
| add intermediate supervision to warped images (ablation) | Kalantari et al. · HCI | PSNR: around 0.3 db | Improved |
| use optimized sampling pattern over neighboring patterns (ablation) | Ours · 12 selected scenes | best PSNR pattern: 9 out of 12 | Improved |
| increase minimum distance of sampling pattern (ablation) | Ours · random sampling patterns | PSNR | Hurt |
| use our method instead of Kalantari et al. with flexible input | Ours · synthetic and real LF datasets | PSNR: around 4 dB on synthetic; 0.4-1 dB on real-world | Improved |
Closest papers by meaning
- Fast Light Field Reconstruction With Deep Coarse-To-Fine Modeling of Spatial-Angular Clues (ECCV 2018 · similarity 0.95)
- Sheared Epipolar Focus Spectrum for Dense Light Field Reconstruction (TPAMI 2023 · similarity 0.90)
- Light Field Reconstruction via Deep Adaptive Fusion of Hybrid Lenses (TPAMI 2023 · similarity 0.90)
- Learning Dynamic Interpolation for Extremely Sparse Light Fields With Wide Baselines (ICCV 2021 · similarity 0.90)
- S2D-LFE: Sparse-to-Dense Light Field Event Generation (CVPR 2025 · similarity 0.87)
- TurboSL: Dense Accurate and Fast 3D by Neural Inverse Structured Light (CVPR 2024 · similarity 0.86)
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.
