Trained as what you said but got bad performance

[shuluoshu]

Hi, @laughtervv
I trained as what you suggested, first pretrain the segmentation model, then use only SIM loss for the first 5 epoch and finally use the total loss. The total loss dropped to 20 and stays after 100 epochs. However, when I do the prediction, the results are bad: mAP : 0.004, other data is also far away from what you gives in your paper. I wonder what happens. The loss of yours can be as low as how much?

Thanks.

[laughtervv]

0.004 looks like a bug or so. Can you try the pretrained model?
Can I see your training log and 'pergroup_thres.txt' and 'mingroupsize.txt'?

[shuluoshu]

log.txt

The above is log.txt, and next is the pergroup_thres.txt and mingroupsize.txt

pergroup_thres.txt

mingroupsize.txt

Thanks @laughtervv

[laughtervv]

There's something wrong with your 'pergroup_thres.txt'.
Here are the files I generated. Please give it a try.
pergroup_thres.txt
mingroupsize.txt

[jianuo1128]

@laughtervv @shuluoshu
Hello author, I used your pre-trained model, and when running the valid.py, I encountered the following nan problem, and the two txt files generated are as follows