HRNet
The parallel multi-resolution backbone that reset pose accuracy records at CVPR 2019.
About
HRNet changed how pose backbones are designed: instead of downsampling and then recovering resolution, it keeps a high-resolution stream alive through the whole network, adding lower-resolution streams in parallel and fusing them repeatedly. The result is position-sensitive features that are unusually good at precise keypoint localization.
It broke three COCO records on release and became the default backbone inside HigherHRNet, HybrIK, and a generation of top-down pose models. The TPAMI journal version generalized the idea beyond pose to classification, detection, and segmentation. Accuracy-wise it has since been passed, but the design is still a fixture in the field.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-reported (CVPR 2019 paper): 77.7 AP on COCO test-dev with the W48 backbone; widely used as the baseline other models compare against, but no single independent benchmark of HRNet itself as a product was 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 (verified, metrics-v1). Evidence: peer-reviewed CVPR 2019 paper and TPAMI journal extension by named MSRA/USTC researchers; foundational and enormously cited; code and weights publicly available. Strengths: strong provenance, foundational influence, real code and weights. Docked for: (1) I could not verify the exact license of the official release in this pass; (2) superseded in raw accuracy by newer models. License: not verified — treat weights as research-use pending confirmation.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 10.0/10
- Maintenance · recency, releases, issue load
- 5.0/10
- Openness · license + docs
- 6.0/10
- Manual review · sanity read + red-flag check
- 9.0/10
- · CVPR 2019 (Sun, Xiao, Liu, Wang; MSRA/USTC); parallel multi-resolution design, TPAMI journal extension.
- · Provider-claimed 77.7 AP COCO test-dev (W48); foundational, still a default backbone (HigherHRNet, HybrIK).
- · Accuracy since surpassed by newer models.