EvalRef3D

Record raw model performance on independent holdout subjects.

Validation runs locally in your browser. Only Ground Truth and raw prediction labelmaps go into the Evaluation Request ZIP. Source MRI/image voxels and model weights stay on your computer.

A Model Evaluation Certificate is a self-evaluated research record. It is not a performance guarantee or clinical validation.

1. Select Model ZIP

One original TrainRef3D Model ZIP

2. Add Holdout Training ZIPs

SegRef3D Training Data ZIPs with complete Ground Truth

3. Add InferRef3D Result ZIPs

Original raw results from the selected model

Select all cases together (1–32 cases, request up to 1 GiB). File names do not determine pairing. No resampling, postprocessing or manual corrections are accepted.

4. Pair & Validate

5. Confirm the evaluation set

Evaluated data domain

Record modality, population and acquisition context. Case count and domain are part of every certificate.

I confirm that these evaluation cases are independent subjects that were not used for training or internal validation, and that the Ground Truth annotations are complete for the selected target.

Subject independence is user-attested. The platform cannot detect the same subject exported under different IDs.

6. Create Evaluation Request ZIP

7. Open EvalRef3D Colab

Open EvalRef3D Colab ↗

Use a CPU runtime. Upload the downloaded request explicitly. Opening Colab uploads nothing.

Select one model and the holdout archives.