The size of the gt is mismatch with the size of the results.

[djl0912111]

Sorry to bother you. I have just found that the results in DUTS-TE have several pictures mismatch. for example, the size in the picture which named ILSVRC2012_test_00036002.png in the DUTS-TE-MASK is 400266, however, corresponding one in the run-0-sal-d which named ILSVRC2012_test_00036002_sal_fuse.png have the size of 400308. Because I want to calculate the MAE, the problem above restrict me, and I have found it is not the only one in the results file. Hope your reply, thank you.

[backseason]

You can resize the predicted map to the same size as the gt. The performance will be nearly the same.

[djl0912111]

Thanks a lot. And I have an other pproblem in evaluation. I follows your repo in other issues getting an url for evaluation,like this https://github.com/Pchank/caffe-sal. However, I found the maxF which calculated is different with the results in the paper. I think it may calculate the average F-measure. Do you have any other code for Max F-measure. Sorry to bother you. Thanks a lot. Hope your reply. Best wishes!

[backseason]

You can use the code mentioned here: https://github.com/backseason/PoolNet#update , which should produce the same evaluation results as reported in the paper.

[djl0912111]

Thanks a lot. Actually, I have tried many times, both in windows and in linux. I still can't install it. Did you evaluate the results using this mode? Did I have any other choice. It 's very kind of you,I really bother you a lot.o(╥﹏╥)o

[backseason]

Yes, that's the code that I've been using. My environment for evaluation is Ubuntu 16(18).04 and Python 2.7.

[djl0912111]

I’ll try it in python2.7 ! Thank you very much! Peace & Love!