Darknet stopping after calculating mAP

Question

I'm currently facing a weird issue. I have recently trained a couple of different yolov4 models using the following command: darknet.exe detector train datapath cfgpath yolov4.conv.137 -map. Usually this works fine, but out of nowhere it now began to stop after each map calculation. Normally it just continues and displays the calculated map to me, but now I need to rerun the same command again (with the last weights instead of yolov4.conv.137) to continue training. Thanks a lot in advance.

Answer 1

I have the same issue once. It only occurs on one of my models.

I guess the problem is due to the memory issue when there're too many images in the list to calculate. I have over 100K images in my training set. So I disable the argument "-map" only when training this model. It also reduces the time needed to train because the calculation just took so long. After all, you can simply calculate the mAP after training process is mature.