Minor edits

app.py

@@ -27,6 +27,7 @@ def preprocess(vidpath):
 
         frames.append(img)
     
+    # frames = torch.cuda.FloatTensor(np.asarray(frames, dtype=np.float32)) if torch.cuda.is_available() else torch.Tensor(np.asarray(frames, dtype=np.float32))
     frames = torch.Tensor(np.asarray(frames, dtype=np.float32))
     
     transform = transforms.Compose([videotransforms.CenterCrop(224)])
@@ -44,28 +45,34 @@ def classify(video,dataset='WLASL100'):
     input = preprocess(video)
 
     model = InceptionI3d()
-    model.cpu()
-    model.load_state_dict(torch.load('weights/rgb_imagenet.pt'))
+    model.load_state_dict(torch.load('weights/rgb_imagenet.pt',map_location=torch.device('cpu')))
     model.replace_logits(to_load[dataset]['logits'])
-    model.load_state_dict(torch.load(to_load[dataset]['path']))
+    model.load_state_dict(torch.load(to_load[dataset]['path'],map_location=torch.device('cpu')))
+
+    # device = torch.device('cuda:0') if torch.cuda.is_available() else torch.device('cpu')
+    # model.to(device)
+    model.cpu()
     model.eval()
 
     with torch.no_grad():
         per_frame_logits = model(input)
     
+    per_frame_logits.cpu()
+    model.cpu()
+
     predictions = rearrange(per_frame_logits,'1 j k -> j k')
     predictions = torch.mean(predictions, dim = 1)
 
     top = torch.argmax(predictions).item()
     _, index = torch.topk(predictions,10)
-    index = index.numpy()
+    index = index.cpu().numpy()
 
     with open('wlasl_class_list.txt') as f:
         idx2label = dict()
         for line in f:
             idx2label[int(line.split()[0])]=line.split()[1]
 
-    predictions = torch.nn.functional.softmax(predictions, dim=0).numpy()
+    predictions = torch.nn.functional.softmax(predictions, dim=0).cpu().numpy()
 
     return {idx2label[i]:float(predictions[i]) for i in index}