Research map / Video / Video segmentation and tracking
Recurrent Back-Projection Network for Video Super-Resolution
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 recurrent back-projection to integrate separate context frames with the target-frame SISR path | RBPN · SPMCS-32 | PSNR: 29.85 -> 31.40 | Helped |
| Add the recurrent decoder and temporal connections to per-frame MISR back-projection (ablation) | RBPN · SPMCS-32 | PSNR: 30.89 -> 31.40 | Helped |
| Use DBPN for SISR and ResNet for MISR instead of using the same module type for both (ablation) | RBPN/2 · SPMCS-32 | PSNR: 30.54 -> 30.96; 30.74 -> 30.96 | Helped |
| Use both past and future context rather than past-only context (ablation) | RBPN/6 · SPMCS-32 | PSNR: 31.64 -> 31.74 | Helped |
| Randomize the order of past context frames (ablation) | RBPN/4 · SPMCS-32 | PSNR: 31.40 -> 31.39 | No clear effect |
| Remove optical-flow input (ablation) | RBPN/5 · SPMCS-32 | PSNR: 31.54 -> 31.36 | Hurt |
| Use only the last HR feature map instead of concatenating the maps (ablation) | RBPN/2 · SPMCS-32 | PSNR: 30.96 -> 30.89 | Hurt |
| Replace concatenation of HR feature maps with an LSTM (ablation) | RBPN/6 · SPMCS-32 | PSNR: 31.64 -> 31.46 | Hurt |
| Increase SISR module depth to six stages (ablation) | RBPN/6 · SPMCS-32 | PSNR: 31.64 -> 31.58 | Hurt |
| Reduce MISR, residual, and decoder depth to three blocks (ablation) | RBPN/6 · SPMCS-32 | PSNR: 31.64 -> 31.39 | Hurt |
| Add bicubic residual learning to predict an image residual (ablation) | RBPN/6 · SPMCS-32 | PSNR: 31.64 -> 31.57 | Hurt |
| Use longer video context (ablation) | RBPN · SPMCS-32 | PSNR: +0.2 dB | Helped |
Closest papers by meaning
- Frame-Recurrent Video Super-Resolution (CVPR 2018 · similarity 0.93)
- Frame and Feature-Context Video Super-Resolution (AAAI 2019 · similarity 0.91)
- Image Super-Resolution via Dual-State Recurrent Networks (CVPR 2018 · similarity 0.90)
- Real-Time Video Super-Resolution With Spatio-Temporal Networks and Motion Compensation (CVPR 2017 · similarity 0.90)
- Deep Back-Projection Networks for Super-Resolution (CVPR 2018 · similarity 0.90)
- One-Step Diffusion for Detail-Rich and Temporally Consistent Video Super-Resolution (NeurIPS 2025 · similarity 0.87)
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.