Research map / Video / Video segmentation and tracking
Deep Blind Video Decaptioning by Temporal Aggregation and Recurrence
Dahun Kim, Sanghyun Woo, Joon-Young Lee, In So Kweon
Check 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 |
|---|---|---|---|
| Use a 3D encoder with a 2D decoder rather than a 2D encoder-2D decoder (ablation) | BVDNet · ChaLearn 2018 LAP Inpainting Track2 validation | PSNR: 33.6803 -> 34.1029 | Improved |
| Use a 3D encoder-2D decoder rather than a 3D encoder-3D decoder (ablation) | BVDNet · ChaLearn 2018 LAP Inpainting Track2 validation | PSNR: 28.4590 -> 34.1029 | Improved |
| Add gradient L1 loss (ablation) | 3D encoder-2D decoder with L1 loss · ChaLearn 2018 LAP Inpainting Track2 validation | PSNR: 34.1029 -> 34.2251 | Improved |
| Add SSIM loss alongside L1 and gradient L1 (ablation) | 3D encoder-2D decoder · ChaLearn 2018 LAP Inpainting Track2 validation | PSNR: 34.2251 -> 34.6544 | Improved |
| Add recurrence stream and temporal consistency constraints (ablation) | BVDNet · ChaLearn 2018 LAP Inpainting Track2 validation, 500 clips | Temporal error: 0.00117 -> 0.00090 | Improved |
| Add recurrence stream (ablation) | BVDNet · ChaLearn 2018 LAP Inpainting Track2 validation | PSNR: 34.6544 -> 34.7055 | Improved |
| Add GAN loss (ablation) | BVDNet · ChaLearn 2018 LAP Inpainting Track2 validation | PSNR: 34.7055 -> 31.2257 | Hurt |
| Remove residual image shortcut (ablation) | BVDNet · ChaLearn 2018 LAP Inpainting Track2 validation | PSNR: 34.7055 -> 34.3892 | Hurt |
| Use 3 input frames instead of 5 (ablation) | BVDNet · ChaLearn 2018 LAP Inpainting Track2 validation | PSNR: 34.7055 -> 33.7895 | Hurt |
| Use 7 input frames instead of 5 (ablation) | BVDNet · ChaLearn 2018 LAP Inpainting Track2 validation | PSNR: 34.7055 -> 34.5063 | Hurt |
| Use 9 input frames instead of 5 (ablation) | BVDNet · ChaLearn 2018 LAP Inpainting Track2 validation | PSNR: 34.7055 -> 34.6260 | Hurt |
Closest papers by meaning
- Recurrent Temporal Aggregation Framework for Deep Video Inpainting (TPAMI 2019 · similarity 0.93)
- Learning Blind Video Temporal Consistency (ECCV 2018 · similarity 0.87)
- Deep Video Inpainting (CVPR 2019 · similarity 0.87)
- Blind Bitstream-corrupted Video Recovery via Metadata-guided Diffusion Model (CVPR 2025 · similarity 0.85)
- Extending Video Masked Autoencoders to 128 frames (NeurIPS 2024 · similarity 0.85)
- Generative Omnimatte: Learning to Decompose Video into Layers (CVPR 2025 · similarity 0.84)
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.
