TrainRef3D

Train a custom 3D segmentation model from SegRef3D Training Data ZIP files.

Training cases are validated locally in your browser.
Nothing is uploaded until you explicitly upload the Dataset ZIP to your own Google Colab runtime.

Already have a Dataset ZIP? Open TrainRef3D Colab ↗

1. Add training cases

Drop SegRef3D Training ZIP files here

Select Training ZIP files

One ZIP per independent case · 1 scalar or 3 RGB channels · 64 cases / 1 GiB per dataset maximum

No cases loaded
Case ID	Source format	Channels	Shape	Spacing (mm)	Labels	Intensity policy	Status	Remove

2. Choose one target structure

One model = one structure. The selected Obj ID becomes foreground; all other labels become background.

Target structure Target name in the trained model

I confirm that the selected target has been completely annotated in every included case.
If the target is absent from a labelmap, it will be treated as a true negative/background case.
選択した構造について、全症例でannotationが完了していることを確認してください。対象labelがない症例はnegative caseとして扱われます。

Ready to validate cases locally.

TrainRef3D

1. Add training cases

2. Choose one target structure

Dataset ready