change tif to png

calculate_background.py

@@ -9,7 +9,7 @@
 ):
     for name in files:
         className = path.split("/")[-1]
-        if ".tif" in name:
+        if ".png" in name:
             img = np.asarray(Image.open(os.path.join(path, name)))
             cumPer += np.sum(img == 0) / (img.shape[0] * img.shape[1])
             num += 1

main_pseudo.py

@@ -810,7 +810,7 @@ def save_pil_image(img, path):
                     ).squeeze(0)
                     # print(pseudo_mask.shape)
 
-                    name = f'{filenames[j].split("/")[1]}.tif'
+                    name = f'{filenames[j].split("/")[1]}.png'
                     save_pil_image(
                         pseudo_mask.detach().cpu().numpy().astype(np.int16),
                         os.path.join(args.prediction_path, name),

part_model/dataloader/part_imagenet.py

@@ -125,7 +125,7 @@ def _get_data(self):
                 filenames = sorted([f.strip() for f in fns.readlines()])
             images.extend([f"{img_path}/{f}.JPEG" for f in filenames])
             masks.extend(
-                [f'{part_path}/{f.split("/")[1]}.tif' for f in filenames]
+                [f'{part_path}/{f.split("/")[1]}.png' for f in filenames]
             )
             labels.extend([l] * len(filenames))
         labels = torch.tensor(labels, dtype=torch.long)

part_model/dataloader/part_imagenet_corrupt.py

@@ -143,8 +143,12 @@ def _get_data(self):
             with open(f"{self.path}/{label}.txt", "r") as fns:
                 filenames = sorted([f.strip() for f in fns.readlines()])
             for severity in range(1, 6):
-                images.extend([f"{img_path}/{f}_{severity}.JPEG" for f in filenames])
-                masks.extend([f'{part_path}/{f.split("/")[1]}.tif' for f in filenames])
+                images.extend(
+                    [f"{img_path}/{f}_{severity}.JPEG" for f in filenames]
+                )
+                masks.extend(
+                    [f'{part_path}/{f.split("/")[1]}.png' for f in filenames]
+                )
                 labels.extend([l] * len(filenames))
         labels = torch.tensor(labels, dtype=torch.long)
         return images, labels, masks
@@ -197,7 +201,9 @@ def get_loader_sampler(args, transform, split, distributed_sampler=True):
     # TODO: can we make this cleaner?
     PART_IMAGENET_CORRUPT["part_to_class"] = part_imagenet_dataset.part_to_class
     PART_IMAGENET_CORRUPT["num_classes"] = part_imagenet_dataset.num_classes
-    PART_IMAGENET_CORRUPT["num_seg_labels"] = part_imagenet_dataset.num_seg_labels
+    PART_IMAGENET_CORRUPT[
+        "num_seg_labels"
+    ] = part_imagenet_dataset.num_seg_labels
 
     setattr(args, "num_classes", part_imagenet_dataset.num_classes)
     pto = part_imagenet_dataset.part_to_object
@@ -231,7 +237,9 @@ def load_part_imagenet(args):
         ]
     )
 
-    train_loader, train_sampler = get_loader_sampler(args, train_transforms, "train")
+    train_loader, train_sampler = get_loader_sampler(
+        args, train_transforms, "train"
+    )
     val_loader, _ = get_loader_sampler(args, val_transforms, "val")
     test_loader, _ = get_loader_sampler(args, val_transforms, "test")
 

part_model/dataloader/part_imagenet_geirhos.py

@@ -123,7 +123,9 @@ def _get_data(self):
             with open(f"{self.path}/{label}.txt", "r") as fns:
                 filenames = sorted([f.strip() for f in fns.readlines()])
             images.extend([f"{img_path}/{f}.png" for f in filenames])
-            masks.extend([f'{part_path}/{f.split("/")[1]}.tif' for f in filenames])
+            masks.extend(
+                [f'{part_path}/{f.split("/")[1]}.png' for f in filenames]
+            )
             labels.extend([l] * len(filenames))
         labels = torch.tensor(labels, dtype=torch.long)
         return images, labels, masks
@@ -173,8 +175,12 @@ def get_loader_sampler(args, transform, split, distributed_sampler=True):
     )
 
     # TODO: can we make this cleaner?
-    PART_IMAGENET_GEIRHOS["part_to_class"] = part_imagenet_geirhos_dataset.part_to_class
-    PART_IMAGENET_GEIRHOS["num_classes"] = part_imagenet_geirhos_dataset.num_classes
+    PART_IMAGENET_GEIRHOS[
+        "part_to_class"
+    ] = part_imagenet_geirhos_dataset.part_to_class
+    PART_IMAGENET_GEIRHOS[
+        "num_classes"
+    ] = part_imagenet_geirhos_dataset.num_classes
     PART_IMAGENET_GEIRHOS[
         "num_seg_labels"
     ] = part_imagenet_geirhos_dataset.num_seg_labels
@@ -211,7 +217,9 @@ def load_part_imagenet_geirhos(args):
         ]
     )
 
-    train_loader, train_sampler = get_loader_sampler(args, train_transforms, "train")
+    train_loader, train_sampler = get_loader_sampler(
+        args, train_transforms, "train"
+    )
     val_loader, _ = get_loader_sampler(args, val_transforms, "val")
     test_loader, _ = get_loader_sampler(args, val_transforms, "test")
 

part_model/dataloader/part_imagenet_imagenet_class.py

@@ -272,7 +272,7 @@ def _get_data(self):
                 filenames = sorted([f.strip() for f in fns.readlines()])
             images.extend([f"{img_path}/{f}.JPEG" for f in filenames])
             masks.extend(
-                [f'{part_path}/{f.split("/")[1]}.tif' for f in filenames]
+                [f'{part_path}/{f.split("/")[1]}.png' for f in filenames]
             )
             labels.extend([l] * len(filenames))
         labels = torch.tensor(labels, dtype=torch.long)

part_model/dataloader/part_imagenet_pseudo.py

@@ -126,7 +126,7 @@ def _get_data(self):
                 filenames = sorted([f.strip() for f in fns.readlines()])
             images.extend([f"{img_path}/{f}.JPEG" for f in filenames])
             masks.extend(
-                [f'{part_path}/{f.split("/")[1]}.tif' for f in filenames]
+                [f'{part_path}/{f.split("/")[1]}.png' for f in filenames]
             )
             labels.extend([l] * len(filenames))
             files.extend(filenames)

part_model/dataloader/part_imagenet_pseudo_imagenet_class.py

@@ -273,7 +273,7 @@ def _get_data(self):
                 filenames = sorted([f.strip() for f in fns.readlines()])
             images.extend([f"{img_path}/{f}.JPEG" for f in filenames])
             masks.extend(
-                [f'{part_path}/{f.split("/")[1]}.tif' for f in filenames]
+                [f'{part_path}/{f.split("/")[1]}.png' for f in filenames]
             )
             labels.extend([l] * len(filenames))
             files.extend(filenames)

post_prediction.py

@@ -96,7 +96,7 @@
 tran_val_count = 0
 for path, subdirs, files in os.walk(old_dataset_path):
     for name in files:
-        if ".tif" in name:
+        if ".png" in name:
             if "train" in path or "val" in path:
                 tran_val_count += 1
             count += 1
@@ -119,7 +119,7 @@
 mask_names = []
 for root, dirs, files in os.walk(prediction_path):
     for file in files:
-        if ".tif" in file:
+        if ".png" in file:
             mask_names.append(file.split(".")[0])
 
 # Step 4
@@ -156,16 +156,16 @@
         # symlink original masks from old dataset
         for fileName in step_all_list:
             os.symlink(
-                f"{relative_path_old_to_new}/{partition}/{c}/{fileName}.tif",  # calculate relative path
-                f"{new_dataset_path}/{partition}/{c}/{fileName}.tif",
+                f"{relative_path_old_to_new}/{partition}/{c}/{fileName}.png",  # calculate relative path
+                f"{new_dataset_path}/{partition}/{c}/{fileName}.png",
             )
 
         # Step 6
         # copy newly generated masks over
         for fileName in partition_generated[partition]:
             shutil.copyfile(
-                f"{prediction_path}/{fileName}.tif",
-                f"{new_dataset_path}/{partition}/{c}/{fileName}.tif",
+                f"{prediction_path}/{fileName}.png",
+                f"{new_dataset_path}/{partition}/{c}/{fileName}.png",
             )
         # write .txt file
         with open(f"{new_dataset_path}/{partition}/{c}.txt", "w") as f:

post_prediction_imagenet.py

@@ -76,7 +76,7 @@
 tran_val_count = 0
 for path, subdirs, files in os.walk(old_dataset_path):
     for name in files:
-        if ".tif" in name:
+        if ".png" in name:
             if "train" in path or "val" in path:
                 tran_val_count += 1
             count += 1
@@ -101,7 +101,7 @@
 mask_names = []
 for root, dirs, files in os.walk(prediction_path):
     for file in files:
-        if ".tif" in file:
+        if ".png" in file:
             mask_names.append(file.split(".")[0])
 
 print(len(mask_names), len(set(mask_names)))
@@ -160,20 +160,20 @@
     # symlink
     for fileName in train_lines:
         os.symlink(
-            f"{relative_path_old_to_new}/train/{c}/{fileName}.tif",  # calculate relative path
-            f"{new_dataset_path}/train/{c}/{fileName}.tif",
+            f"{relative_path_old_to_new}/train/{c}/{fileName}.png",  # calculate relative path
+            f"{new_dataset_path}/train/{c}/{fileName}.png",
         )
     for fileName in val_lines:
         os.symlink(
-            f"{relative_path_old_to_new}/val/{c}/{fileName}.tif",  # calculate relative path
-            f"{new_dataset_path}/train/{c}/{fileName}.tif",
+            f"{relative_path_old_to_new}/val/{c}/{fileName}.png",  # calculate relative path
+            f"{new_dataset_path}/train/{c}/{fileName}.png",
         )
 
     # copy newly generated masks over
     for fileName in class_train_masks:
         shutil.copyfile(
-            f"{prediction_path}/{fileName}.tif",
-            f"{new_dataset_path}/train/{c}/{fileName}.tif",
+            f"{prediction_path}/{fileName}.png",
+            f"{new_dataset_path}/train/{c}/{fileName}.png",
         )
 
 # Step 6
@@ -192,8 +192,8 @@
         # copy newly generated masks over
         for fileName in new_partition_list:
             shutil.copyfile(
-                f"{prediction_path}/{fileName}.tif",
-                f"{new_dataset_path}/{partition}/{c}/{fileName}.tif",
+                f"{prediction_path}/{fileName}.png",
+                f"{new_dataset_path}/{partition}/{c}/{fileName}.png",
             )
         # write .txt file
         with open(f"{new_dataset_path}/{partition}/{c}.txt", "w") as f:

pre_prediction.py

@@ -91,7 +91,7 @@
 tran_val_count = 0
 for path, subdirs, files in os.walk(old_dataset_path):
     for name in files:
-        if ".tif" in name:
+        if ".png" in name:
             if "train" in path or "val" in path:
                 tran_val_count += 1
             count += 1
@@ -152,7 +152,7 @@
                 pass
 
 
-# Create .tif file of correct dimensions inside test dir
+# Create .png file of correct dimensions inside test dir
 def save_pil_image(img, path):
     image_path = os.path.join(path)
     pil_img = Image.fromarray(img)
@@ -168,7 +168,7 @@ def save_pil_image(img, path):
         tif = np.zeros(img.size)
         print(img.size)
         save_pil_image(
-            tif, f"{new_temp_mask_dataset_path}/test/{c}/{name[:-1]}.tif"
+            tif, f"{new_temp_mask_dataset_path}/test/{c}/{name[:-1]}.png"
         )
 
 # Step 6

pre_prediction_imagenet.py

@@ -85,7 +85,7 @@
 tran_val_count = 0
 for path, subdirs, files in os.walk(old_dataset_path):
     for name in files:
-        if ".tif" in name:
+        if ".png" in name:
             if "train" in path or "val" in path:
                 tran_val_count += 1
             count += 1
@@ -143,7 +143,7 @@
             ) as f:
                 pass
 
-# Create .tif file of correct dimensions inside test dir
+# Create .png file of correct dimensions inside test dir
 def save_pil_image(img, path):
     image_path = os.path.join(path)
     pil_img = Image.fromarray(img)
@@ -159,7 +159,7 @@ def save_pil_image(img, path):
         tif = np.zeros(img.size)
         print(img.size)
         save_pil_image(
-            tif, f"{new_temp_mask_dataset_path}/test/{c}/{name[:-1]}.tif"
+            tif, f"{new_temp_mask_dataset_path}/test/{c}/{name[:-1]}.png"
         )
 
 # Step 6

to_class_specific.py

@@ -51,7 +51,7 @@
 # Step 1
 for path, subdirs, files in os.walk(metaclass_dataset_dir):
     for name in files:
-        if ".tif" in name:
+        if ".png" in name:
             metaclass = path.split("/")[-1]
             imagenet_class = name.split("_")[0]
             metaclass_to_class[metaclass].add(imagenet_class)
@@ -115,7 +115,7 @@ def save_pil_image(img, path):
 for path, subdirs, files in os.walk(metaclass_dataset_dir):
     for name in files:
         className = path.split("/")[-1]
-        if ".tif" in name:
+        if ".png" in name:
             img = np.asarray(Image.open(os.path.join(path, name)))
             imagenet_className = name.split("_")[0]