trained model1

Author: wolverinn

README.md

@@ -32,9 +32,9 @@ We generate our own dataset from YUV test sequences, refer to:
 We use Cross Entropy Loss as loss function. For the four output labels, we calculate the Cross Entropy Loss seperately and then add them together.
 
 ## Validation
-The **loss** of our trained model is: 3.1161
+The **loss** of our trained model is: 3.1049
 
-The **accuracy** of each label predicted is: 65.29%
+The **accuracy** of each label predicted is: 66.12%
 
 The best way to evaluate the model is to integrate the model into the HEVC encoder. I've conceived a pipeline:
 

hevc_encoder_model.pt

@@ -80,7 +80,7 @@ def process_rdcost(ctu_x,ctu_y):
             if len(depth_cost[frame_number][ctu_number]["1"]) == 4:
                 pass
             else:
-                if (int(ctu_number) + 1) % ctu_x == 0 and (int(ctu_number) + 1) // ctu_x == ctu_y:
+                if (int(ctu_number) + 1) % ctu_x == 0 and ((int(ctu_number) + 1) // ctu_x) == ctu_y:
                     depth_cost[frame_number][ctu_number]["0"] = sum(depth_cost[frame_number][ctu_number]["1"])
                     depth_cost[frame_number][ctu_number]["1"] += [0,0,0]
                     depth_cost[frame_number][ctu_number]["2"] += list(np.zeros(12))