Datasets:
| { | |
| "repository": "https://github.com/digitalpathologybern/hover_next_train", | |
| "revision": "db4adfa425ffcad51d01a361e660a40c1ecda9dd", | |
| "classes": { | |
| "0": "background", | |
| "1": "neutrophil", | |
| "2": "epithelial-cell", | |
| "3": "lymphocyte", | |
| "4": "plasma-cell", | |
| "5": "eosinophil", | |
| "6": "connective-tissue-cell", | |
| "7": "mitosis" | |
| }, | |
| "evidence": { | |
| "src/constants.py": "CLASS_NAMES defines the seven foreground names in order.", | |
| "src/data_utils.py": "add_3c_gt_fast reads Y[...,0] as instances; get_weighted_sampler reads gt[...,1] as classes 0..7. get_lizard reads source fold_0 train/valid and separate test arrays.", | |
| "src/post_proc_utils.py": "convert_reg counts object classes == i+1 for class_names[i], establishing one-based foreground mapping." | |
| }, | |
| "source_array_scope": "Only lizard_mitosis.zip, not separate mitosis_ds/eos_val/mit_val archives. Source array index is the only per-image identifier available in this ZIP." | |
| } | |