SpursLipu
SpursLipu
Thank you for your quick reply. My version of Openvino is 2019 R1, and I have already cheak my bashrc file. Just like following picture.  But it still doesn't...
Still doesn't work..... 
 Still doesn't work...... I don't knwo why, and I want to cry.......

I also solved this problem as you said, think you very much!
你把 -pt去掉就可以了。
-pt是在使用coco80类预训练文件的时候使用。mAP训练也无法提升吗?
不会一直很低吧?还是要看最后的best.pt,有突变的很低也正常
转化是darknet weights是可以的,你可以用我的convert.py这个文件。
你可以试试减小batchsize 或者减小imgsize,yolov4就是比较吃显存