problems about high pixel pictures

[Zhousiyuhit]

Hi, if I want to train pictures in 2002003, do you have some advices for me about how to choose the network weights? Maybe I think 2 conv and 2 fc layers are not satisfied. Thank you very much.

[Zhousiyuhit]

sorry, there are some errors about my issue, pictures in 200X200X3.

[a7b23]

Hi i think you should increase the number of conv layers and increase the hidden units in fc layers too.

[Zhousiyuhit]

Yes, but I know nothing about CNN, can you give me some advices about the parameters in hidden layers? How do you choose the right parameters in your networks? I encountered some problems and don’t know how to continue my works. By the way, I want to use DQN to recognize the pictures provided from UAV and send them into UGV to guide UGV to find the most optimal path in Mars. Maybe it sounds like a strange idea.

在 2017年10月18日，下午3:22，a7b23 [email protected] 写道：

Hi i think you should increase the number of conv layers and increase the hidden units in fc layers too.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/a7b23/Autonomous-MazePathFinder-using-DQN/issues/3#issuecomment-337483187, or mute the thread https://github.com/notifications/unsubscribe-auth/AXZNzoZApR6TviVXh8JrvAf_Rnx6xc5Xks5stacngaJpZM4P7fcm.