Open JINXvvv opened 4 years ago
Training with the 300w dataset? The model is small, and it should be efficient. Please debug in the dataiter, findout which image transform delay the training. And it is better to train with a normal pc that with a ssd, other than a server.
我的loss一直停留在14左右降不下去,尝试修改过batchsize,但是没什么效果 而且训练速度特别慢,10个iter要一两分钟,2080ti的显卡利用率只有2% 请问您有什么建议吗