Research map / Video / Video segmentation and tracking
Exploiting Optical Flow Guidance for Transformer-Based Video Inpainting
Kaidong Zhang, Jialun Peng, Jingjing Fu, Dong Liu
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 global tokens to local window attention (ablation) | FGT++ spatial transformer · DAVIS (square masks) | PSNR: 31.37 -> 31.62 | Improved |
| add TD-MHSA (ablation) | FGT++ temporal transformer · DAVIS (square masks) | PSNR: 31.62 -> 31.87 | Improved |
| add FGFP module (ablation) | FGT++ · DAVIS (square masks) | PSNR: 31.62 -> 32.36 | Improved |
| add FGFI module (ablation) | FGT++ · DAVIS (square masks) | PSNR: 31.62 -> 31.82 | Improved |
| combine flow completion with deformable convolution in FGFP (ablation) | FGT++ FGFP · DAVIS (square masks) | PSNR: 31.76 -> 32.36 | Improved |
| use FGFI instead of simple concatenation (ablation) | FGT++ · DAVIS (square masks) | PSNR: 31.68 -> 31.82 | Improved |
| integrate FGFI in more than the first spatial transformer (ablation) | FGT++ · DAVIS (square masks) | PSNR: 31.82 -> 31.87 | No clear effect |
| use 4 TD-MHSA blocks instead of 1 (ablation) | FGT++ · DAVIS (square masks) | PSNR: 31.66 -> 31.87 | Improved |
| add FGFP to blocks beyond encoder+3 (ablation) | FGT++ · DAVIS (square masks) | PSNR: 32.36 -> 32.34 | No clear effect |
| add local aggregation to flow completion (ablation) | LAFC · DAVIS (square masks) | EPE: 0.546 -> 0.524 | Improved |
| add edge loss to flow completion (ablation) | LAFC · DAVIS (square masks) | EPE: 0.524 -> 0.511 | Improved |
| add amplitude loss (ablation) | FGT++ · DAVIS (square masks) | PSNR: 32.62 -> 33.01 | Improved |
Closest papers by meaning
- Flow-Guided Transformer for Video Inpainting (ECCV 2022 · similarity 0.94)
- Towards an End-to-End Framework for Flow-Guided Video Inpainting (CVPR 2022 · similarity 0.91)
- Deep Flow-Guided Video Inpainting (CVPR 2019 · similarity 0.90)
- WaveFormer: Wavelet Transformer for Noise-Robust Video Inpainting (AAAI 2024 · similarity 0.87)
- Video Diffusion Models Are Strong Video Inpainter (AAAI 2025 · similarity 0.85)
- Elevating Flow-Guided Video Inpainting with Reference Generation (AAAI 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.
