search

bodokaiser / piwise

Pixel-wise segmentation on VOC2012 dataset using pytorch.
BSD 3-Clause "New" or "Revised" License
383 stars 86 forks source link

Probability maps instead of Binary maps #6

Closed saeedizadi closed 7 years ago

saeedizadi commented 7 years ago

Hello, In the evaluation phase, your code outpus binary segmentation maps (for a binary segmentation task), while there cases in which a the real-value probability map is desired. How can I change your code to get the probability map?

Thanks Saeed

bodokaiser commented 7 years ago

Hey Saeed,

Check this line here.

You find the probability map of the i-th image of the j-th class with outputs[i].cpu()[j].data. However as with ~21 different classes this is far more difficult to visualize (that's why I use ouputs[i].cpu().max(0)[1].data to get the class with the highest probability for each pixel).

Hope that helps.

saeedizadi commented 7 years ago

Thanks for the kind reply :)