Hello @Spider-scnu, first, thanks for sharing this awesome work on GitHub.
I read the full version of the paper available here, and found out that it references to itself 2 times :
Maybe that's a typo you missed.
Also I was wondering, did you try to make the heatmap initialized with something decreasing with the distance between 2 edges - instead of everything set to 1 as you did - for highlighting the importance of RL ?
Maybe it would have been a fairer comparison to the RL approach. Shimomura, M. and Takashima, Y. said that this kind of method gave them way better result than random only.
Hello @Spider-scnu, first, thanks for sharing this awesome work on GitHub. I read the full version of the paper available here, and found out that it references to itself 2 times : Maybe that's a typo you missed.
Also I was wondering, did you try to make the heatmap initialized with something decreasing with the distance between 2 edges - instead of everything set to 1 as you did - for highlighting the importance of RL ? Maybe it would have been a fairer comparison to the RL approach. Shimomura, M. and Takashima, Y. said that this kind of method gave them way better result than random only.
For example :
(same in white) :
Kind regards,
Clement