Closed enquel closed 2 years ago
Hi enquel,
Thanks for reaching out to us. Does the preprocessed Barracuda input values match the inputs in the original demo?
@Aurimasp I will check asap (which in my case means may take a while; side project), and get back to you.
@enquel on which hardware are you running the inference? And on which backend?
Closing this issue as there is/has been no activity for some time. Please reopen if needed.
Barracuda 2.21, model from here: https://github.com/onnx/models/blob/master/vision/body_analysis/emotion_ferplus/README.md. Resulting classes are always the same with non-significant differences, despite proper preprocessing of images (single channel, 8bit, no mip maps, etc.)