Nanochallenge

Internal challenges: submit an inference image, it runs on held-out vxData cases, and its scores land on the dashboard's Nanochallenge page.

Available Challenges

Challenge	Description	Details
`wsi-gleason-segmentation`	Pixel-level Gleason grading of WSI crops	CHALLENGE.md
`bpmri-cspca-prediction`	csPCa detection on bi-parametric MRI, scored like PI-CAI	CHALLENGE.md
`bpmri-isup-prediction`	Per-case ISUP grade prediction on bi-parametric MRI	CHALLENGE.md
`in-vivo-microscope`	Histology read back out of in-vivo prostate MRI, voxel by voxel	CHALLENGE.md

Submitting to a Challenge

vx challenge list
vx challenge describe -c bpmri-cspca-prediction
vx challenge submit -c bpmri-cspca-prediction -i mono/my-model@sha256:ab12cd... \
  -t "my run" --model v2 --notes "baked weights"   # --key value: free-form metadata

The image must be pinned by digest and implement infer.sh -i /inputs -o /outputs.

Adding a Challenge

Create challenges/<name>/ with a Dockerfile, a pixi.toml with prepare and evaluate tasks (evaluate via challenge_utils.scoring.run, so the dashboard can show the result), and a CHALLENGE.md.
Add its entry (name, title, description, primary_metric) to the index annotation in argo/nanochallenge.workflowtemplate.yaml; a dashboard-api test checks every challenge has one.
Build its image, always mono/nanochallenge-<name>: ./build.sh challenges/<name>/Dockerfile (./build.sh challenges builds all).
Add a row to the table above.