The difference between test error and test error with p-cutoff?

Hello all,

Who knows the difference between test error and test error with p-cutoff ? Which one is more suitable for evaluating the performance of training dataset?

Thanks so much!

If you have occlusions, then p-cutoff is more meaningful, as otherwise the pixel error is confounded with completely “off” points; I would report both and what your p-cutoff is in a methods section of a paper.

1 Like

Thanks so much for the explanation!

1 Like