FastSCNN has two outputs

[butnaruteodor]

hello,

I just trained a FastSCNN model on a custom dataset and exported it to onnx, now i'm wondering why does this model have two outputs? shouldn't there be only one? and if i have these two outputs could i somehow fuse them to get better segmentation.

I attached an image with the outputs read by netron, the model was simplified using onnx-simplifier