The reproduced results are relatively poor

[liuxingyu123]

Hello, I reproduced the code according to your steps, but the training results are ODS: 0.755, OIS: 0.773, AP: 0.650, which is different from your results. What's the problem?

[xwjabc]

Hmmm that is strange. So you simply evaluate my pretrained version (see the command below) and got the low values?

python hed.py --checkpoint ./data/hed_checkpoint.pt --output ./output-mypretrain --test
cd eval
(echo "data_dir = '../output-mypretrain/test'"; cat eval_edge.m)|matlab -nodisplay -nodesktop -nosplash

[liuxingyu123]

your pretrained result is OIS:0.788,ODS:0.808,AP:0.840，but my reproduced result is lower,Is it because of the environment? My environment is pytorch1.7, python3.7.

[xwjabc]

I see. It is possible due to the environment issue. My previous environment is PyTorch 1.0 + Python 3.6. Let me try to reproduce it with PyTorch 1.7.0 + Python 3.7. Stay tuned.

[liuxingyu123]

ok,thank you very much.

[liuxingyu123]

hello,i try to reproduce it with pytorch1.0+python3.6 , the result is same on pytorch1.6+python3.7 lower than your OIS:0.788,ODS:0.808,AP:0.840,i follow readme to reproduce .do you have other idea to fix this problem? thank you.

[liuxingyu123]

The tested plot diagram is abnormal. Do you know why

[liuxingyu123]

I solved this problem, I found my eval code is wrong, thank you very much.