Help with inference/post-processing for the scrfd model for face detection

@omria Hi, tagging you as you seem knowledgable in this area.

My network outputs for the face landmarks are in UINT8, but I need them to be converted to float32 and dequantized properly - but I don’t have the dequantization values since I didn’t quantize the model myself.

Related post: Where to quantize inputs - #5 by omria