test graph optimizations

app.py

@@ -51,6 +51,14 @@ def sigmoid(x):
     return 1 / (1 + np.exp(-x))
 
 
+def create_transform(img_size):
+    return transforms.Compose([
+        SquarePad(),
+        transforms.Resize((img_size, img_size), interpolation=Image.BICUBIC),
+        transforms.ToTensor(),
+    ])
+
+
 def inference(stream_url, start_time, end_time, count_only_api, api_key, 
               img_size=288, seq_len=64, stride_length=32, stride_pad=3, batch_size=4, 
               miss_threshold=0.8, marks_threshold=0.5, median_pred_filter=True, center_crop=True, both_feet=True, 
@@ -63,6 +71,7 @@ def inference(stream_url, start_time, end_time, count_only_api, api_key,
         providers = [("CUDAExecutionProvider", {"device_id": torch.cuda.current_device(),
                                                 "user_compute_stream": str(torch.cuda.current_stream().cuda_stream)})]
         sess_options = ort.SessionOptions()
+        sess_options.graph_optimization_level = ort.GraphOptimizationLevel.ORT_ENABLE_ALL
         ort_sess = ort.InferenceSession(onnx_file, sess_options=sess_options, providers=providers)
     else:
         ort_sess = ort.InferenceSession(onnx_file)
@@ -105,27 +114,12 @@ def inference(stream_url, start_time, end_time, count_only_api, api_key,
         all_frames.append(all_frames[-1])
     batch_list = []
     idx_list = []
+    preprocess = create_transform(img_size)
     for i in tqdm(range(0, length + stride_length - stride_pad, stride_length)):
         batch = all_frames[i:i + seq_len]
         Xlist = []
         print('Preprocessing...')
         for img in batch:
-            transforms_list = []
-            # if center_crop:
-            #     if width > height:
-            #         transforms_list.append(transforms.Resize((int(width / (height / img_size)), img_size)))
-            #     else:
-            #         transforms_list.append(transforms.Resize((img_size, int(height / (width / img_size)))))
-            #     transforms_list.append(transforms.CenterCrop((img_size, img_size)))
-            # else:
-            transforms_list.append(SquarePad())
-            transforms_list.append(transforms.Resize((img_size, img_size), interpolation=Image.BICUBIC))
-            
-
-            transforms_list += [
-                transforms.ToTensor()]
-                #transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])]
-            preprocess = transforms.Compose(transforms_list)
             frameTensor = preprocess(img).unsqueeze(0)
             Xlist.append(frameTensor)