problem about your lstm block

[milk-bottle-liyu]

Hi, I read your code and found a problem about your lstm block. Usually we don't shuffle the input data of lstm, but in your provider model, you shuffle the file names in function provider.read_from().

Won't shuffle break the data sequence or I miss something?
Thank u~

[wangjksjtu]

Hi, @public1024. You are correct! When you run the lstm model, the data sequence has to be maintained. You need to comment this line like the provider (Provider2) for testing.

DBNet/provider.py

Line 103 in 53dd1b2

random.shuffle(c)

Note that this demo code is only tested on the nvidia-pn model. You are welcome to submit pull-requests for other settings :) Recently, I am busy in doing other things. I will provide full supports as well as the results if I get a chance.

[milk-bottle-liyu]

thanks! Wonderful job!