Help with implementing semantic segmentation in three ways

If Semantic segmentation is implemented in below three ways using BCEWithLogitsLoss and Dice score metric, will it give same outputs?
Assume Input Image Dimension [Batch,Channels,height,width] & number of classes is 21(VOC).

a) Converting the labels into one hot encoding where target label Dimension is [Batch,Num_classes,height,width].
b) Converting the labels into categorical values where target label Dimension is [Batch,1,height,width]. implemented repo https://github.com/mapillary/inplace_abn
c) Converting the labels into RGB values where target label Dimension is [Batch,3,height,width] implemented repo https://github.com/jfzhang95/pytorch-deeplab-xception

For all this methods whether prediction output shape should be [Batch,Channels,height,width] ?