Why do you use 0.1 as the threshold?

ArcherFMY / sal_eval_toolbox

evaluation toolbox for salient object detection

MIT License

147 stars 40 forks source link

Why do you use 0.1 as the threshold? #5

Closed lartpang closed 5 years ago

lartpang commented 5 years ago

https://github.com/ArcherFMY/sal_eval_toolbox/blob/8e3837f783cea892e13a47601a1329836c7860f0/tools/Curve_BenchCode/performCalcu.m#L42

The DUTS's ground truth is not entirely binary.

Thanks....

ArcherFMY commented 5 years ago

The basic code is from my senior colleagues. They used this threshold to binarize the ground truth. I just tried the 0.5 as the threshold and there is no big difference when evaluating DUTS-test dataset.

I think it is ok as long as we evaluate all algorithms using the same settings.

lartpang commented 5 years ago

OK, thanks!