Update app.py

app.py

@@ -65,32 +65,6 @@ def write_flo(flow, filename):
     f.close()
 
 
-def warp_flow(img, flow, mul=1.):
-    #img = np.array(img.convert('RGB'))  
-    img = cv2.imread(img)
-    flow = cv2.imread(flow)
-    #flow = np.load(flow)
-    h, w = flow.shape[:2]
-    flow = flow.copy()
-    flow[:, :, 0] + np.arange(w)
-    flow[:, :, 1] + np.arange(h)[:, np.newaxis]
-    # print('flow stats', flow.max(), flow.min(), flow.mean())
-    # print(flow)
-    flow*mul
-    # print('flow stats mul', flow.max(), flow.min(), flow.mean())
-    # res = cv2.remap(img, flow, None, cv2.INTER_LINEAR)
-    res = cv2.remap(img, flow, None, cv2.INTER_LANCZOS4)
-    
-    print(res)
-
-def get_warp_res(fname_image, fname_flow, fname_output='warped.png'):
-    print(f"FNAME IMAGE: {fname_image}")
-    #im2 = imread(fname_image)
-    #print(f"FNAME IMAGE READED: {im2.shape}")
-    #flow = fname_flow.cpu().detach().numpy()
-    flow = fname_flow
-    #print(f"FNAME FLOW READED: {flow.shape}")
-    res = warp_flow(fname_image, flow, 1.)
     
 def infer():
     video_url = "https://download.pytorch.org/tutorial/pexelscom_pavel_danilyuk_basketball_hd.mp4"
@@ -183,17 +157,10 @@ def infer():
     print(f"predicted flow shape = {predicted_flow.shape}")
     
     flow_img = flow_to_image(predicted_flow).to("cpu")
-#     output_folder = "/tmp/"  # Update this to the folder of your choice
     write_jpeg(flow_img, f"predicted_flow.jpg")
     
-    #input_image = flow_to_image(frames[100]).to("cpu")
-    #write_jpeg(input_image, f"frame_input.jpg")
-    
     flo_file = write_flo(predicted_flow, "flofile.flo")
-    #write_jpeg(frames[100], f"input_image.jpg")
-    #res = warp_image(img1_batch, predicted_flow)
-    
-    
+      
     # define a transform to convert a tensor to PIL image
     transform = T.ToPILImage()
     
@@ -202,15 +169,12 @@ def infer():
     img = img.resize((960, 520))
     # display the PIL image
     #img.show()
-    frame1pil = np.array(img.convert('RGB'))
-    print(f"frame1pil: {frame1pil}")
-    print(f"frame1pil shape: {frame1pil.shape}")
-    print(f"frame1pil dtype: {frame1pil.dtype}")
-    img.save('frame_input.jpg')
-    print(f"img1_batch: {img1_batch}")
-    print(f"img1_batch shape: {img1_batch.shape}")
-    print(f"img1_batch dtype: {img1_batch.dtype}")
-    #res = get_warp_res(img1_batch, "predicted_flow.jpg", 'warped.png')
+    frame2pil = np.array(img.convert('RGB'))
+    print(f"frame1pil: {frame2pil}")
+    print(f"frame1pil shape: {frame2pil.shape}")
+    print(f"frame1pil dtype: {frame2pil.dtype}")
+    img.save('raw_frame2.jpg')
+    
 
     numpy_array_flow = predicted_flow.permute(1, 2, 0).detach().cpu().numpy()
     print(f"numpy_array_flow: {numpy_array_flow}")
@@ -226,17 +190,17 @@ def infer():
     numpy_array_flow*1.
     # print('flow stats mul', flow.max(), flow.min(), flow.mean())
     # res = cv2.remap(img, flow, None, cv2.INTER_LINEAR)
-    res = cv2.remap(frame1pil, numpy_array_flow, None, cv2.INTER_LANCZOS4)
+    res = cv2.remap(frame2pil, numpy_array_flow, None, cv2.INTER_LANCZOS4)
     print(res)
     
     res = Image.fromarray(res)
     res.save('wraped.jpg')
 
     blend2 = Image.open('frame_input.jpg')
-    image4 = Image.blend(res,blend2,0.5)
-    image4.save("output2.jpg")
+    blend2 = Image.blend(res,blend2,0.5)
+    blend2.save("blended2.jpg")
     
-    return "done", "predicted_flow.jpg", ["flofile.flo"], 'frame_input.jpg', 'wraped.jpg', "output2.jpg"
+    return "done", "predicted_flow.jpg", ["flofile.flo"], 'raw_frame2.jpg', 'wraped.jpg', "blended2.jpg"
 ####################################
 # Bonus: Creating GIFs of predicted flows
 # ---------------------------------------