# ufsm super-resolution checkpoints: srp1
x4 CT super-resolution (ufsm `train --task restore --sr-pairs 1`, https://github.com/SuperOptimizer/ufsm): native ESRF scans at
8.64 / 9.36 um -> ~2.4 um grid, trained on real registered pairs (native scan -> the same scroll's 2.2-2.4 um scan resampled through
our registrations, per-pair monotone intensity map onto the native scan's intensities). Residual net on the x4 trilinear upsampling.
Use: `ufsm predict <ckpt> <root> <native.zarr> <out.zarr> --um <native um>`. latest.json: step, validation (PSNR / SSIM / gradient PSNR
per pair after the intensity map), training configuration. Only the newest 3 checkpoints are kept here.