Research map / 3D vision / Depth and stereo
Light Field Reconstruction Using Convolutional Network on EPI and Extended Applications
Gaochang Wu, Yebin Liu, Lu Fang, Qionghai Dai, Tianyou Chai
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 blur and deblur steps around the CNN (ablation) | Our CNN · Real-world 30 scenes | PSNR: 37.15 -> 41.02 | Improved |
| add blur and deblur steps around the CNN (ablation) | Our CNN · Stanford Lytro Reflective 29 | PSNR: 44.84 -> 46.10 | Improved |
| add blur and deblur steps around the CNN (ablation) | Our CNN · Stanford Lytro Occlusion 16 | PSNR: 35.89 -> 38.86 | Improved |
| add learned CNN restoration to blur-bicubic-deblur (ablation) | blur-deblur framework · Real-world 30 scenes | PSNR: 34.97 -> 41.02 | Improved |
| add learned restoration instead of bicubic (ablation) | blur-deblur framework · HCI Buddha 3x3 | PSNR: 34.71 -> 43.20 | Improved |
| add learned restoration instead of bicubic (ablation) | blur-deblur framework · HCI Buddha 5x5 | PSNR: 39.23 -> 46.42 | Improved |
| add learned restoration instead of bicubic (ablation) | blur-deblur framework · HCI Mona 3x3 | PSNR: 38.21 -> 44.37 | Improved |
| add learned restoration instead of bicubic (ablation) | blur-deblur framework · HCI Mona 5x5 | PSNR: 42.66 -> 51.07 | Improved |
| use our residual network instead of FSRCNN as restoration step (ablation) | blur-deblur framework · HCI Mona 5x5 | PSNR: 49.75 -> 51.07 | Improved |
| use our residual network instead of FSRCNN as restoration step (ablation) | blur-deblur framework · HCI Buddha 3x3 | PSNR: 42.60 -> 43.20 | Improved |
| use Gaussian blur kernel instead of sinc kernel (ablation) | EPI blur step · Stanford Light Field Archive | MSE between blurred and upsampled EPIs and blurred ground truth: 0.153 -> 0.061 | Improved |
| use Gaussian blur kernel instead of Butterworth kernel (ablation) | EPI blur step · Stanford Light Field Archive | MSE between blurred and upsampled EPIs and blurred ground truth: 0.089 -> 0.061 | Improved |
Closest papers by meaning
- Light Field Reconstruction Using Deep Convolutional Network on EPI (CVPR 2017 · similarity 0.97)
- Neural EPI-Volume Networks for Shape From Light Field (ICCV 2017 · similarity 0.91)
- Residual Networks for Light Field Image Super-Resolution (CVPR 2019 · similarity 0.91)
- High-Dimensional Dense Residual Convolutional Neural Network for Light Field Reconstruction (TPAMI 2019 · similarity 0.91)
- EPINET: A Fully-Convolutional Neural Network Using Epipolar Geometry for Depth From Light Field Images (CVPR 2018 · similarity 0.90)
- Learning Non-Local Spatial-Angular Correlation for Light Field Image Super-Resolution (ICCV 2023 · 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.
