Overall context:
(1) Resize the input picture
(2) Feature extraction network extracts feature layer
(3) Obtain prediction results from features
(4) Use the prediction results of the network to adjust the prior frame to get the final prediction frame-draw it up
(5) Training after painting
Feature extraction network extracts feature layer
1. The prediction part-the prediction result is the change of the prior frame
1. Obtain prediction results from features
2. Decoding of prediction results
3. Draw on the original image
Second, the training part
1. The processing of the real frame
2. Calculate Loss using the processed real frame and the prediction result of the corresponding picture