The problem with evaluation.

[badbadcode]

Thanks for your work.

If you binarize the prediction results at a threshold of 0.5 and save them as {0, 255},

UACANet/run/Test.py

Line 66 in 6b9e54e

Image.fromarray(((out['pred'] > .5) * 255).astype(np.uint8)

what is the significance of setting different thresholds during evaluation?

UACANet/run/Eval.py

Line 94 in 6b9e54e

for j, threshold in enumerate(Thresholds):