About ade20k_loader

[Eniac-Xie]

Hi, thank you for your code!

I want to use the ptsemseg/loader/ade20k_loader.py, in the function:

    def encode_segmap(self, mask):
        # Refer : http://groups.csail.mit.edu/vision/datasets/ADE20K/code/loadAde20K.m
        mask = mask.astype(int)
        label_mask = np.zeros((mask.shape[0], mask.shape[1]))
        label_mask = ( mask[:,:,0] / 10.0 ) * 256 + mask[:,:,1]
        return np.array(label_mask, dtype=np.uint8)

the label_mask is with elements that larger than 256, so why do you use dtype=np.uint8 (which means mask[:,:,0] / 10.0 ) * 256 will be discarded).

[meetps]

n_classes is only 150, so I think the max value label_mask can take is 150 (inside uint8 range).

Can you point to a specific instance from the ADE20K dataset where this doesn't work?

[jxlijunhao]

@meetshah1995 @Eniac-Xie I have same problem and confused by this encode utils, for exampe, the images in the trainning set a/abbey/ADE_train_00000971.jpg

using above functions and np.unique(mask), the results will gave:
[ 0 39 56 101 116 255]