uber-research / UPSNet

UPSNet: A Unified Panoptic Segmentation Network
Other
649 stars 120 forks source link

Pan output exceeds number of classes #132

Open discretecoder opened 4 years ago

discretecoder commented 4 years ago

I trained the network on 38 classes. but when i test the model, i tried printing the np.unique(pan) from the output and it has class ids greater than 38 so when i try to feed the sem to color generator, it throws an error since there are only 38 classes in the gt. how do i fix this

following is how the np.unique(pan) looks like. all the unique pans after 37 do not make any sense as these classes do not exist in the training [ 2 3 4 10 14 15 18 20 22 24 25 26 27 28 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 255] [ 2 3 4 10 14 15 18 20 22 24 25 26 27 28 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 255] [ 2 3 4 9 10 14 18 20 22 23 24 25 26 27 28 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 255] [ 2 3 4 10 14 18 20 22 23 24 25 26 27 28 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 255]