add plotting of ROCs

Author: Duftata

evaluation/quantitative.py

@@ -6,7 +6,7 @@
 
 from dataset.grazpedwri_dataset import GrazPedWriDataset
 
-mode = ['end2end', 'lin_eval'][1]
+mode = ['end2end', 'lin_eval'][0]
 
 metrics_kwargs = {'num_labels': GrazPedWriDataset.N_CLASSES, 'average': None}
 metrics = MetricCollection({

visualisation/plot_aurocs.py

@@ -0,0 +1,43 @@
+from pathlib import Path
+
+import torch
+from matplotlib import pyplot as plt
+from torchmetrics import classification
+
+from dataset.grazpedwri_dataset import GrazPedWriDataset
+
+models2plot = ['image', 'image_frac_loc', 'image_frac_loc_bin_seg_clip']
+
+pred_dir = Path('evaluation/predictions')
+gt = torch.load(pred_dir / 'ground_truth.pt')
+
+path_dict = {k.stem.rsplit('_', 1)[0]: k for k in pred_dir.iterdir() if not k.is_dir()}
+pred_dict = dict()
+for experiment in models2plot:
+    pred = torch.load(path_dict[experiment])
+    y = []
+    y_hat = []
+    for file_stem in gt.keys():
+        y.append(gt[file_stem])
+        y_hat.append(pred[file_stem])
+    y = torch.stack(y).int()
+    y_hat = torch.stack(y_hat)
+    pred_dict[experiment] = y_hat
+
+
+roc = classification.MultilabelROC(num_labels=len(models2plot))
+for c in range(GrazPedWriDataset.N_CLASSES):
+    cat_y_hat = torch.stack([pred_dict[e][:, c] for e in models2plot], dim=1)
+
+    roc.update(cat_y_hat, y[:, c].unsqueeze(1).expand(-1, len(models2plot)))
+    fig, axs = roc.plot(score=True, labels=models2plot)
+    axs.set_title('')
+    plt.legend(fontsize='large')
+
+    fig.savefig(f'/home/ron/Documents/Konferenzen/BVM 2025/ROCs/roc_{GrazPedWriDataset.CLASS_LABELS[c].replace('/', '_')}.pdf',
+                bbox_inches='tight', pad_inches=0)
+
+    roc.reset()
+plt.show()
+
+