Pre-trained model weights for YADO (You Accurately Denoise real Observations) Denoiser and Renoising DM (ECCV 2026)

Model weights for YADO (You Accurately Denoise real Observations) Denoiser and Renoiser as introduced in the paper "Rethinking Real-World MRI Denoising: Learning from Physical Noise" accepted to ECCV 2026. See also the Project Page.

The code and docs are at: github.com/Deep-MI/YADO, and the weights are mostly also compatible with YODA. The checkpoints each contain the weights (`ckpt/last.pth`) and the `config.yml`. 

Model recommendation: Choose by contrast first, then use the (best-)matching resolution. Always use guided (`g-*`) models when corresponding co-acquired contrast is available (at high resolution). At test time prefer ReN2N over pN2N as we have consistently seen better generalization, unless your data matches the protocol of the respective training dataset.

Note: Just try out it out! We were ourselves surprised to see how well YADO generalizes, e.g. on Tumor data (incl contrast-enhanced T1w) or when applying the HCP-A (u-ReN2N; 3T ME-MPRAGE @ 0.8 mm) to 7T MP2RAGE @ 7T.

Denoiser (YADO) checkpoints

Renoising DM (ReN) checkpoints