Research map / Video / Video segmentation and tracking
An Efficient Recurrent Adversarial Framework for Unsupervised Real-Time Video Enhancement
Dario Fuoli, Zhiwu Huang, Danda Pani Paudel, Luc Van Gool, Radu Timofte
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 LGM (local-global) modules to recurrent RaGAN* model (ablation) | RAVE · Vid3oC | FID: 59.10 -> 51.85 | Improved |
| add LGM modules to non-recurrent RaGAN* model (ablation) | RAVE · Vid3oC | FID: 60.63 -> 53.37 | Improved |
| add recurrence to RaGAN* model (ablation) | RAVE · Vid3oC | FID: 60.63 -> 59.10 | Improved |
| add recurrence alone to baseline GAN (ablation) | baseline GAN · Vid3oC | FVD: 1048.91 -> 1092.26 | Hurt |
| add RaGAN* objective to baseline GAN (ablation) | baseline GAN · Vid3oC | FID: 60.91 -> 60.63 | Improved |
| add LGM in full model on runtime (ablation) | RAVE · Vid3oC | Runtime [ms]: 25.3 -> 27.8 | Hurt |
| use single joint discriminator instead of two discriminators (ablation) | RAVE · Vid3oC | FID: 51.25 -> 51.85 | No clear effect |
| vary LGM global channel size from 8 to 48 (ablation) | RAVE · Vid3oC | FID: 51.85 -> 54.85 | Hurt |
| vary LGM global channel size from 8 to 4 (ablation) | RAVE · Vid3oC | FID: 51.85 -> 52.20 | Hurt |
| use RAVE instead of single-frame enhancer WESPE | WESPE · Huawei | FID: 115.84 -> 113.54 | Improved |
| use RAVE instead of DPE for speed | DPE · FullHD runtime | Runtime [ms]: 117.9 -> 27.8 | Improved |
| use RAVE instead of video translation method | video translation baseline · Vid3oC | FID: 56.68 -> 51.85 | Improved |
Closest papers by meaning
- Learning Blind Video Temporal Consistency (ECCV 2018 · similarity 0.86)
- Video Dynamics Prior: An Internal Learning Approach for Robust Video Enhancements (NeurIPS 2023 · similarity 0.86)
- OSV: One Step is Enough for High-Quality Image to Video Generation (CVPR 2025 · similarity 0.86)
- Diffusion Adversarial Post-Training for One-Step Video Generation (ICML 2025 · similarity 0.85)
- FreeLong: Training-Free Long Video Generation with SpectralBlend Temporal Attention (NeurIPS 2024 · similarity 0.85)
- Autoregressive Adversarial Post-Training for Real-Time Interactive Video Generation (NeurIPS 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.
