Change names to image, target and output

Author: synml

demo.py

@@ -37,21 +37,21 @@
     groundtruth_dir = os.path.join('demo', 'groundtruth')
     os.makedirs(result_dir, exist_ok=True)
     os.makedirs(groundtruth_dir, exist_ok=True)
-    for images, masks in tqdm.tqdm(testloader, desc='Demo'):
+    for image, target in tqdm.tqdm(testloader, desc='Demo'):
         # mask에 255를 곱하여 0~1 사이의 값을 0~255 값으로 변경 + 채널 차원 제거
-        masks.mul_(255).squeeze_(dim=1)
+        target.mul_(255).squeeze_(dim=1)
 
-        images, masks = images.to(device), masks.type(torch.LongTensor)
+        image, target = image.to(device), target.type(torch.LongTensor)
 
         # 예측
         with torch.no_grad():
-            masks_pred = model(images)
-            masks_pred = F.log_softmax(masks_pred, dim=1)
-            masks_pred = torch.argmax(masks_pred, dim=1)
+            output = model(image)
+            output = F.log_softmax(output, dim=1)
+            output = torch.argmax(output, dim=1)
 
         # 1 배치단위 처리
-        assert masks.shape[0] == masks_pred.shape[0]
-        for i in range(masks.shape[0]):
-            plt.imsave(os.path.join(result_dir, image_names[step]), masks_pred[i].cpu(), cmap=cmap)
-            plt.imsave(os.path.join(groundtruth_dir, image_names[step]), masks[i], cmap=cmap)
+        assert target.shape[0] == output.shape[0]
+        for i in range(target.shape[0]):
+            plt.imsave(os.path.join(result_dir, image_names[step]), output[i].cpu(), cmap=cmap)
+            plt.imsave(os.path.join(groundtruth_dir, image_names[step]), target[i], cmap=cmap)
             step += 1

eval.py

@@ -48,27 +48,27 @@ def evaluate(model, testloader, criterion, num_classes: int, device):
     metrics = EvaluationMetrics(num_classes)
     val_loss = 0
     inference_time = 0
-    for images, masks in tqdm.tqdm(testloader, desc='Eval', leave=False):
+    for image, target in tqdm.tqdm(testloader, desc='Eval', leave=False):
         # mask에 255를 곱하여 0~1 사이의 값을 0~255 값으로 변경 + 채널 차원 제거
-        masks.mul_(255).squeeze_(dim=1)
+        target.mul_(255).squeeze_(dim=1)
 
-        images, masks = images.to(device), masks.to(device, dtype=torch.int64)
+        image, target = image.to(device), target.to(device, dtype=torch.int64)
 
         # 예측
         with torch.no_grad():
             start_time = time.time()
-            masks_pred = model(images)
+            output = model(image)
             inference_time += time.time() - start_time
 
         # validation loss를 모두 합침
-        val_loss += criterion(masks_pred, masks).item()
+        val_loss += criterion(output, target).item()
 
         # Segmentation map 만들기
-        masks_pred = F.log_softmax(masks_pred, dim=1)
-        masks_pred = torch.argmax(masks_pred, dim=1)
+        output = F.log_softmax(output, dim=1)
+        output = torch.argmax(output, dim=1)
 
         # 혼동행렬 업데이트
-        metrics.update_matrix(masks, masks_pred)
+        metrics.update_matrix(target, output)
 
     # 평가 점수 가져오기
     iou, miou = metrics.get_scores(ignore_first_label=True)

feature_visualizer.py

@@ -70,7 +70,7 @@ def hook(model, input, output):
 
     # 예측
     with torch.no_grad():
-        mask_pred = model(image)
+        output = model(image)
 
     # 각 계층의 feature maps 저장
     for layer in tqdm.tqdm(feature_maps.keys(), desc='Saving'):

train.py

@@ -37,16 +37,16 @@
     for epoch in tqdm.tqdm(range(config[config['model']]['epoch']), desc='Epoch'):
         model.train()
 
-        for batch_idx, (images, masks) in enumerate(tqdm.tqdm(trainloader, desc='Train', leave=False)):
+        for batch_idx, (image, target) in enumerate(tqdm.tqdm(trainloader, desc='Train', leave=False)):
             # mask에 255를 곱하여 0~1 사이의 값을 0~255 값으로 변경 + 채널 차원 제거
-            masks.mul_(255).squeeze_(dim=1)
+            target.mul_(255).squeeze_(dim=1)
 
-            images, masks = images.to(device), masks.to(device, dtype=torch.int64)
+            image, target = image.to(device), target.to(device, dtype=torch.int64)
 
             # 순전파 + 역전파 + 최적화
             optimizer.zero_grad()
-            masks_pred = model(images)
-            loss = criterion(masks_pred, masks)
+            output = model(image)
+            loss = criterion(output, target)
             loss.backward()
             optimizer.step()