EDVR
The NTIRE 2019 video restoration champion that swept all four tracks with deformable alignment — the ancestor of modern VSR.
About
EDVR (CVPRW 2019, arXiv 1905.02716) is the video restoration framework from Xintao Wang, Kelvin Chan, and colleagues that won all four tracks of the NTIRE 2019 video restoration and enhancement challenges — the first NTIRE edition to tackle video after three years of image-only contests. Its two signature modules are the Pyramid, Cascading and Deformable (PCD) alignment module, which aligns frames at the feature level with deformable convolutions in a coarse-to-fine cascade to survive large motions, and the Temporal-Spatial Attention (TSA) fusion module, which weights the aligned features before reconstruction.
EDVR is a unified framework: the same architecture handles super-resolution, deblurring, and denoising, and the repo also ships the lighter EDVR-M variant plus training code. It seeded the BasicSR ecosystem and the whole deformable-alignment lineage that BasicVSR++ later refined. It is dated by today's standards, but as a foundational, fully open implementation it remains a reliable starting point and teaching reference.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Competition result (NTIRE 2019, CVPRW): first place in all four video restoration/enhancement tracks, ahead of second place by a large margin; author-reported SOTA vs published methods on REDS at release
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: NTIRE 2019 sweep documented by CUHK (ie.cuhk.edu.hk) and the CVPRW paper arXiv 1905.02716; official repo xinntao/EDVR with test and training code. Strengths: competition-verified win (independent judging, all four tracks), foundational open implementation, BasicSR ecosystem support. Docked for: (1) 2019-era method, superseded by BasicVSR++/transformers on quality; (2) repo itself is archival with no active maintenance; (3) quality claims beyond the competition are author-reported. 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
- 4.0/10
- Openness · license + docs
- 8.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · CVPRW 2019, arXiv 1905.02716 (Wang et al., CUHK MMLab + NTU/SIAT/SenseTime)
- · NTIRE 2019: first place in all four video restoration/enhancement tracks (independently judged)
- · PCD alignment + TSA fusion; EDVR-M lightweight variant; BasicSR ecosystem
- · 2019-era; superseded on quality by BasicVSR++/transformers; archival repo