Datasets
Dataset images are gitignored (large binary files). This directory is populated locally or via the artifact store.
Versions
| Directory | Description |
|---|---|
seg_training/ |
Primary segmentation training set |
2label_annotated_v1/ |
Images from 1st training with 2 labels |
4label_reannotated_v1/ |
Images from 1st training, reannotated with 4 labels |
Splits
Train/val/test manifests are in data/splits/ and only save the label file name. Images should be created as links to
the datasets such that it saves the space and we keep only one copy of the overall image entries.