Docs recommend using tf1 when converting to onnx from tf/keras. Is that still a thing? Asking with hope that it might be outdated docs.
Looks like tf2onnx does now support exporting TF-2.x to onnx.
I know at one point they were having stability issues, but now it seems ok. So I would go with that.
Let me know
1 Like
Everything went super smoothly. Ran into a minor issue but that’s all on me. Didn’t realize the Tensor(texture) constructor outputs 0-1 values. I rescale within the model as well.
Thanks for the hard work on barracuda <3
awesome, glad to hear it