Research map / Image generation / Super-resolution
Improved Adversarial Diffusion Compression for Real-World Video Super-Resolution
Bin Chen, Weiqi Li, Shijie Zhao, Xuanyu Zhang, Junlin Li, Li zhang, Jian Zhang
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 1D temporal convolutions to the 2D backbone (versus 3D DiT teacher size) (ablation) | AdcVSR · UDM10/YouHQ40 and efficiency | parameters and E_warp: narrows DISTS gap to 3D model to 0.0014 with 7% parameters | Improved |
| add temporal modeling (1D convs) to the 2D SD backbone AdcSR (ablation) | AdcVSR · Real-VSR benchmarks | video quality and E_warp | Improved |
| double the number of 1D convolutions (ablation) | AdcVSR · Real-VSR benchmarks | video quality | No clear effect |
| replace 1D convolutions with temporal self-attention (ablation) | AdcVSR · Real-VSR benchmarks | quality and loss | No clear effect |
| use dual heads instead of a single head (with dual-domain) (ablation) | AdcVSR discriminator · YouHQ40 | E_warp: 6.32 -> 2.22 | Improved |
| add pixel-domain discriminator to dual-head feature-domain (ablation) | AdcVSR discriminator · YouHQ40 | CLIPIQA: 0.6421 -> 0.6861 | Improved |
| add adversarial loss to distillation (ablation) | AdcVSR · MVSR4x | LPIPS: 0.3596 -> 0.3337 | Improved |
| add DOVE teacher guidance versus GT only (ablation) | AdcVSR · MVSR4x | MUSIQ: 50.32 -> 61.48 | Improved |
| use DOVE as teacher instead of SeedVR2 (ablation) | AdcVSR · MVSR4x | LPIPS: 0.3489 -> 0.3337 | Improved |
| use DOVE as teacher instead of DLoRAL (ablation) | AdcVSR · MVSR4x | MUSIQ: 54.61 -> 61.48 | Improved |
| share backbone and tail convolutions between detail and consistency heads (ablation) | AdcVSR discriminator · training stability | training convergence | Improved |
| add shuffled-video and assembled-image negatives to head training data (ablation) | AdcVSR dual-head · Real-VSR | E_warp and LPIPS | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DOVE — DOVE: Efficient One-Step Diffusion Model for Real-World Video Super-Resolution (NeurIPS 2025) · better on 6 of 9 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Adversarial Diffusion Compression for Real-World Image Super-Resolution (CVPR 2025 · similarity 0.94)
- TEASR: Training-Efficient Any-Step Diffusion Transformer for Real-World Image Super-Resolution (ECCV 2026 · similarity 0.91)
- FlashVSR: Towards Real-time Diffusion-Based Streaming Video Super Resolution (CVPR 2026 · similarity 0.90)
- DOVE: Efficient One-Step Diffusion Model for Real-World Video Super-Resolution (NeurIPS 2025 · similarity 0.90)
- Unleashing the Power of One-Step Diffusion based Image Super-Resolution via a Large-Scale Diffusion Discriminator (NeurIPS 2025 · similarity 0.88)
- Noise-Started One-Step Image Super-Resolution via LR-Conditioned SplitMeanFlow and GAN Refinement (NeurIPS 2026 · 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.
