Hi,
Thanks for providing useful tools.
I want to calculate the ROC but for almost 4000000000 (4B) images. Do you have any suggestion to optimize the code?
Thanks for reading.

I am getting very good segmentation performance with my custom dataset, however, when I tried to plot the precision recall curves and ROC curves, the ROC curve turned out to be good but the MAP value was the least, however, my precision recall curve was very close to ideal. I have attached the PR graph [herewith:]

)
Is there an issue with the MAP formula used in the autometric code?
MAP = np.round(np.sum((Precision[1:] + Precision[:-1]) * (Recall[:-1] - Recall[1:])) / 2.,4)

Thanks for your help.

What's groundtruth image and predict probabilistic image?
Could you give me an example? Thanks!