fix liznerski#8 and added runner for liznerski#15

- No error is raised anymore if no anomalous samples are available during testing for global normalized heatmaps generation (liznerski#8)
- Added a script that loads a snapshot and provides pixel-wise anomaly score tensors for a folder of images (liznerski#15)

python/fcdd/runners/run_prediction_with_snapshot.py

@@ -0,0 +1,85 @@
+import torch
+import torchvision.transforms as transforms
+from torch.utils.data.dataloader import DataLoader
+from fcdd.training.fcdd import FCDDTrainer
+from fcdd.models.fcdd_cnn_224 import FCDD_CNN224_VGG_F
+from fcdd.datasets.image_folder import ImageFolder
+from fcdd.datasets.preprocessing import local_contrast_normalization
+
+# -------- MvTec-AD pre-computed min and max values per class after lcn1 has been applied, copied from fcdd.datasets.mvtec --------
+min_max_l1 = [
+    [(-1.3336724042892456, -1.3107913732528687, -1.2445921897888184),
+     (1.3779616355895996, 1.3779616355895996, 1.3779616355895996)],
+    [(-2.2404820919036865, -2.3387579917907715, -2.2896201610565186),
+     (4.573435306549072, 4.573435306549072, 4.573435306549072)],
+    [(-3.184587001800537, -3.164201259613037, -3.1392977237701416),
+     (1.6995097398757935, 1.6011602878570557, 1.5209171772003174)],
+    [(-3.0334954261779785, -2.958242416381836, -2.7701096534729004),
+     (6.503103256225586, 5.875098705291748, 5.814228057861328)],
+    [(-3.100773334503174, -3.100773334503174, -3.100773334503174),
+     (4.27892541885376, 4.27892541885376, 4.27892541885376)],
+    [(-3.6565306186676025, -3.507692813873291, -2.7635035514831543),
+     (18.966819763183594, 21.64590072631836, 26.408710479736328)],
+    [(-1.5192601680755615, -2.2068002223968506, -2.3948357105255127),
+     (11.564697265625, 10.976534843444824, 10.378695487976074)],
+    [(-1.3207964897155762, -1.2889339923858643, -1.148416519165039),
+     (6.854909896850586, 6.854909896850586, 6.854909896850586)],
+    [(-0.9883341193199158, -0.9822461605072021, -0.9288841485977173),
+     (2.290637969970703, 2.4007883071899414, 2.3044068813323975)],
+    [(-7.236185073852539, -7.236185073852539, -7.236185073852539),
+     (3.3777384757995605, 3.3777384757995605, 3.3777384757995605)],
+    [(-3.2036616802215576, -3.221003532409668, -3.305514335632324),
+     (7.022546768188477, 6.115569114685059, 6.310940742492676)],
+    [(-0.8915618658065796, -0.8669204115867615, -0.8002046346664429),
+     (4.4255571365356445, 4.642300128936768, 4.305730819702148)],
+    [(-1.9086798429489136, -2.0004451274871826, -1.929288387298584),
+     (5.463134765625, 5.463134765625, 5.463134765625)],
+    [(-2.9547364711761475, -3.17536997795105, -3.143850803375244),
+     (5.305514812469482, 4.535006523132324, 3.3618252277374268)],
+    [(-1.2906527519226074, -1.2906527519226074, -1.2906527519226074),
+     (2.515115737915039, 2.515115737915039, 2.515115737915039)]
+]
+# ---------------------------------------------------------------------------------------------------------------------------------
+
+
+# Path to your snapshot.pt
+snapshot = "fcdd/data/mvtec_snapshot.pt"
+
+# Pick the architecture that was used for the snapshot (mvtec's architecture defaults to the following)
+net = FCDD_CNN224_VGG_F((3, 224, 224), bias=True).cuda()
+
+# Path to a folder that contains a subfolder containing the images (this is required to use PyTorch's ImageFolder dataset).
+# For instance, if the images are in foo/my_images/xxx.png, point to foo. Make sure foo contains only one folder (e.g., my_images).
+images_path = "fcdd/data/datasets/foo"
+
+# Pick the class the snapshot was trained on.
+normal_class = 0
+
+# Use the same test transform as was used for training the snapshot (e.g., for mvtec, per default, the following)
+transform = transforms.Compose([
+    transforms.Resize(224),
+    transforms.CenterCrop(224),
+    transforms.ToTensor(),
+    transforms.Lambda(lambda x: local_contrast_normalization(x, scale='l1')),
+    transforms.Normalize(
+        min_max_l1[normal_class][0],
+        [ma - mi for ma, mi in zip(min_max_l1[normal_class][1], min_max_l1[normal_class][0])]
+    )
+])
+
+# Create a trainer to use its loss function for computing anomaly scores
+ds = ImageFolder(images_path, transform, transforms.Lambda(lambda x: 0))
+loader = DataLoader(ds, batch_size=16, num_workers=0)
+trainer = FCDDTrainer(net, None, None, (None, None), None, 'fcdd', 8, 0.99, 128)  # these parameters will have no effect (used for heatmaps)
+trainer.load(snapshot)
+all_anomaly_scores = []
+for inputs, labels in loader:
+    inputs = inputs.cuda()
+    with torch.no_grad():
+        outputs = trainer.net(inputs)
+        anomaly_scores = trainer.anomaly_score(trainer.loss(outputs, inputs, labels, reduce='none'))
+        anomaly_scores = trainer.net.receptive_upsample(anomaly_scores, reception=True, std=8, cpu=False)
+        all_anomaly_scores.append(anomaly_scores.cpu())
+
+# all_anomaly_scores will be a tensor containing pixel-wise anomaly scores for all images
+all_anomaly_scores = torch.cat(all_anomaly_scores)

python/fcdd/training/bases.py

@@ -486,12 +486,11 @@ def _create_heatmaps_picture(self, idx: List[int], name: str, inpshp: torch.Size
         rows = []
         for s in range(number_of_rows):
             rows.append(self._image_processing(imgs[idx][s * nrow:s * nrow + nrow], inpshp, maxres=self.resdown, qu=1))
-            err = self.objective != 'ae' and 'train' not in name  # training samples might have just one label
             if self.objective != 'hsc':
                 rows.append(
                     self._image_processing(
                         ascores[idx][s * nrow:s * nrow + nrow], inpshp, maxres=self.resdown, qu=self.quantile,
-                        colorize=True, ref=balance_labels(ascores, labels, err) if norm == 'global' else ascores[idx],
+                        colorize=True, ref=balance_labels(ascores, labels, False) if norm == 'global' else ascores[idx],
                         norm=norm.replace('semi_', ''),  # semi case is handled in the line above
                     )
                 )
@@ -500,7 +499,7 @@ def _create_heatmaps_picture(self, idx: List[int], name: str, inpshp: torch.Size
                     self._image_processing(
                         grads[idx][s * nrow:s * nrow + nrow], inpshp, self.blur_heatmaps,
                         self.resdown, qu=self.quantile,
-                        colorize=True, ref=balance_labels(grads, labels, err) if norm == 'global' else grads[idx],
+                        colorize=True, ref=balance_labels(grads, labels, False) if norm == 'global' else grads[idx],
                         norm=norm.replace('semi_', ''),  # semi case is handled in the line above
                     )
                 )
@@ -535,20 +534,19 @@ def _create_singlerow_heatmaps_picture(self, idx: List[int], name: str, inpshp:
         """
         for norm in ['local', 'global']:
             rows = [self._image_processing(imgs[idx], inpshp, maxres=res, qu=1)]
-            err = self.objective != 'ae' and 'train' not in name  # training samples might have just one label
             if self.objective != 'hsc':
                 rows.append(
                     self._image_processing(
                         ascores[idx], inpshp, maxres=res, colorize=True,
-                        ref=balance_labels(ascores, labels, err) if norm == 'global' else None,
+                        ref=balance_labels(ascores, labels, False) if norm == 'global' else None,
                         norm=norm.replace('semi_', ''),  # semi case is handled in the line above
                     )
                 )
             if grads is not None:
                 rows.append(
                     self._image_processing(
                         grads[idx], inpshp, self.blur_heatmaps, res, colorize=True,
-                        ref=balance_labels(grads, labels, err) if norm == 'global' else None,
+                        ref=balance_labels(grads, labels, False) if norm == 'global' else None,
                         norm=norm.replace('semi_', ''),  # semi case is handled in the line above
                     )
                 )