Here's a particularly useful XSeg mask I'd like to share. I hope someone can convert it to ONNX format so that software like Rope can utilize this more effective XSeg.
DFL XSeg.npy
https://drive.google.com/drive/folders/1UQVvBR1tR2HREwX_apQ2sArvWEMRGTNI?usp=drive_link