RealBasicVSR
The real-world sibling of BasicVSR++: a cleaning front-end plus stochastic degradations so the model survives genuine low-quality footage.
About
RealBasicVSR (CVPR 2022, arXiv 2111.12704) is the companion paper to BasicVSR++ from the same NTU team, and it asks the practical question: what happens when the input is not a cleanly downsampled lab video but real degraded footage? The answer is a two-part design. First, a cleaning module scrubs input distortions before the super-resolution network ever sees them. Second, a stochastic degradation scheme synthesizes varied degradations during training so the model does not overfit to any single blur or noise profile. The paper also studies the tradeoffs between long-term temporal propagation and artifact amplification, which is where its title comes from.
The repo carries test and training code, video demos, a Colab, and a Hugging Face Space demo, and it released the VideoLQ dataset of real low-quality clips for evaluation. Like its sibling it is a research artifact with no official API, but it remains the go-to open baseline for real-world video restoration rather than synthetic benchmarks.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Author-reported (CVPR 2022): improved real-world restoration over synthetic-trained baselines on the authors' VideoLQ dataset; no independent benchmark audit found
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (metrics tier, metrics-v1). Evidence: CVPR 2022 paper arXiv 2111.12704 from the NTU team; repo confirms test/training code, Colab and a Hugging Face Space demo added March 2022. Strengths: addresses the genuinely harder real-world degradation problem, open weights and training code, community demos, active downstream use. Docked for: (1) benchmark evidence is author-reported on their own VideoLQ dataset; (2) original repo maintenance is minimal; (3) no official API or commercial path. License: research code; repo license terms unverified.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.0/10
- Maintenance · recency, releases, issue load
- 5.0/10
- Openness · license + docs
- 8.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · CVPR 2022, arXiv 2111.12704 (Chan et al., NTU S-Lab); companion to BasicVSR++
- · Cleaning module + stochastic degradation scheme for real-world footage
- · HF Space demo (akhaliq/RealBasicVSR, March 2022); VideoLQ dataset released
- · Benchmarks author-reported on the authors' own VideoLQ dataset