Support training UNet using segmentation datasets (with polygons rather than masks)
The PR checks if it's a mask-based or polygon-based training experiment using the attached dataset versions' names which we will add as constraints to the models.
mask-based training: "images" and "masks"
polygon/segmentation dataset-based training: 1 "full" containing images with polygon annotations
On Picsellia, we will have two models for UNet: UNet-segmentation and UNet-masks. They will share the same docker image, but the latter has the image and mask prefixes added as parameters, for mask and image matching purposes.
Support training UNet using segmentation datasets (with polygons rather than masks)
The PR checks if it's a mask-based or polygon-based training experiment using the attached dataset versions' names which we will add as constraints to the models.
On Picsellia, we will have two models for UNet: UNet-segmentation and UNet-masks. They will share the same docker image, but the latter has the image and mask prefixes added as parameters, for mask and image matching purposes.