Research map / 3D vision / Depth and stereo
Fast Light-Field Disparity Estimation With Multi-Disparity-Scale Cost Aggregation
Zhicong Huang, Xuemei Hu, Zhou Xue, Weizhu Xu, Tao Yue
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 edge guidance sub-network (ablation) | FastLFnet · 4D Light Field (HCI) | MSE x100: 1.844 -> 1.218 | Improved |
| add BAM attention module (ablation) | FastLFnet · 4D Light Field (HCI) | MSE x100: 1.492 -> 1.218 | Improved |
| use multi-disparity-scale cost aggregation instead of one scale (ablation) | FastLFnet · 4D Light Field (HCI) | MSE x100: 1.650 -> 1.218 | Improved |
| add edge guidance sub-network (cropped evaluation) (ablation) | FastLFnet · 4D Light Field (Training+Stratified, 11px crop), average | MSE x100: 1.928 -> 1.291 | Improved |
| add vertical stream to horizontal-only input (16 channels) (ablation) | FastLFnet · 4D Light Field (HCI) | MSE x100: 2.068 -> 1.218 | Improved |
| add diagonal streams to 2-streams input (16 channels) (ablation) | FastLFnet · 4D Light Field (HCI) | MSE x100: 1.218 -> 1.437 | Hurt |
| raise feature channels from 16 to 32 (2-streams) (ablation) | FastLFnet · 4D Light Field (HCI) | MSE x100: 1.218 -> 1.653 | Hurt |
| raise feature channels from 4 to 16 (2-streams) (ablation) | FastLFnet · 4D Light Field (HCI) | MSE x100: 1.756 -> 1.218 | Improved |
| raise feature channels from 16 to 32 on running time (2-streams) (ablation) | FastLFnet · 4D Light Field (HCI) | running time (s): 0.593 -> 1.438 | Hurt |
| use FastLFnet instead of LFattNet on efficiency | FastLFnet vs LFattNet · 4D Light Field (Training+Stratified, 11px crop) | running time (s): 5.862 -> 0.624 | Improved |
| FastLFnet versus LFattNet on accuracy | FastLFnet vs LFattNet · 4D Light Field (Training+Stratified, 11px crop), average | MSE x100: 0.955 -> 1.291 | Hurt |
| FastLFnet versus DslfNet on sparse light fields | FastLFnet vs DslfNet · Sparse Light Field dataset (4 test scenes), average | MSE: 0.32 -> 0.12 | Improved |
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.
- EpiNet-7 — EPINET: A Fully-Convolutional Neural Network Using Epipolar Geometry for Depth From Light Field Images (CVPR 2018) · better on 3 of 4 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
- Take Your Model Further: A General Post-refinement Network for Light Field Disparity Estimation via BadPix Correction (AAAI 2023 · similarity 0.89)
- Shape Estimation from Shading, Defocus, and Correspondence Using Light-Field Angular Coherence (TPAMI 2016 · similarity 0.89)
- Fast Light-Weight Near-Field Photometric Stereo (CVPR 2022 · similarity 0.88)
- Towards Multimodal Depth Estimation From Light Fields (CVPR 2022 · similarity 0.88)
- Disentangling Light Fields for Super-Resolution and Disparity Estimation (TPAMI 2022 · similarity 0.88)
- SynLF: Zero-Shot Metric Depth from Light Field Cameras via Physics-Grounded Synthesis (ECCV 2026 · similarity 0.88)
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.
