Closed mrka124 closed 6 years ago
Sorry, this thesis is written only in Chinese. The related part about firearm is published as a conference paper at ICPR 2018. You can find it here.
To speed training, you can try to reduce the image size, which is very important. You can also forward multiple images and do one backward. You may find the code here helpful. Specifically, you can check the code in the source file train_retr_from_cls.py
.
Thank you!!!
Do you have English version of your thesis? I have same problem with you. My dataset contain many image with different size, many orientation,.. like firearm.
Also I'm using Pytorch with batch size = 1 (I have SPP layer in network), forward one image at time. So my training time is very slow. Do you have any suggestion?
Sorry for my bad English.