Research map / Video / Video segmentation and tracking
An Internal Learning Approach to Video Inpainting
Haotian Zhang, Long Mai, Ning Xu, Zhaowen Wang, John Collomosse, Hailin Jin
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 |
|---|---|---|---|
| train on whole video instead of frame-wise DIP (ablation) | DIP · Composed dataset (250 videos) | FID: 22.3 -> 16.1 | Improved |
| train on whole video instead of frame-wise DIP (ablation) | DIP · Composed dataset (250 videos) | PSNR: 25.2 -> 28.7 | Improved |
| replace 2D with 3D convolution (ablation) | DIP-Vid · Composed dataset (250 videos) | FID: 16.1 -> 12.1 | Improved |
| replace 2D with 3D convolution (ablation) | DIP-Vid · Composed dataset (250 videos) | PSNR: 28.7 -> 30.9 | Improved |
| add flow prediction with consistency loss to 2D network (ablation) | DIP-Vid · Composed dataset (250 videos) | FID: 16.1 -> 10.4 | Improved |
| add flow prediction with consistency loss, versus 3D convolution (ablation) | DIP-Vid-3DCN · Composed dataset (250 videos) | temporal consistency PSNR: 26.7 -> 28.1 | Improved |
| add flow prediction with consistency loss (ablation) | DIP-Vid · Composed dataset (250 videos) | PSNR: 28.7 -> 32.1 | Improved |
| use internal learning instead of Huang et al. patch-based method (all videos) | Huang et al. · Composed dataset (All) | PSNR: 32.3 -> 32.1 | No clear effect |
| use internal learning instead of Huang et al. on complex-motion videos | Huang et al. · Composed dataset (Complex) | PSNR: 31.6 -> 31.9 | Improved |
| use internal learning instead of Huang et al. on simple videos | Huang et al. · Composed dataset (Simple) | PSNR: 33.0 -> 32.2 | Hurt |
| increase input window length k (ablation) | DIP-Vid-Flow · 10 Composed videos, window length 1 to 100 | inpainting quality (full frame) | Improved |
| increase input window length k (ablation) | DIP-Vid-Flow · 10 Composed videos, window length 1 to 100 | non-hole reconstruction quality | Hurt |
Closest papers by meaning
- Internal Video Inpainting by Implicit Long-Range Propagation (ICCV 2021 · similarity 0.92)
- Deep Learning-Based Image and Video Inpainting: A Survey (IJCV 2024 · similarity 0.91)
- Deep Video Inpainting (CVPR 2019 · similarity 0.90)
- Video Diffusion Models Are Strong Video Inpainter (AAAI 2025 · similarity 0.88)
- Towards Enhanced Image Inpainting: Mitigating Unwanted Object Insertion and Preserving Color Consistency (CVPR 2025 · similarity 0.88)
- Elevating Flow-Guided Video Inpainting with Reference Generation (AAAI 2025 · 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.
