wuhuikai/TF-A2RL

doesnt seems to crop only return image similar to input

Closed this issue · 1 comments

doesnt seems to crop only return image similar to input

The algorithm conducts actions randomly sampled from the predicted probs. Thus, sometimes it just outputs the input image. I suggest running the algorithm multiple times to get the desired output.
Or you can modify the code to select the action with the max prob every time.