hqtang

Hi, Did you find out the reason, I also take about 434 secs for the first epoch in loss_train.backward() by cpu training model.

@ccwss1232 Do you find the difference? I also have the same confusion. You can find the detail derivation from this [url] (https://aipiano.github.io/2019/03/23/%E4%B8%8D%E5%8F%98%E6%89%A9%E5%B1%95%E5%8D%A1%E5%B0%94%E6%9B%BC%E6%BB%A4%E6%B3%A21/)) for the question 2.

@sharish33 Have you trained the network with Kitti dataset without cuda? When I ran the training without cuda , Why the loss_train.backward() token about 434 secs just for one epoch?...