Converting tensorflow, keras models to pytorch/fastai


I am wondering if there is a possible way to interface interoperable ML models (TF, keras, MXNet) to pytorch/fastai. So far I’ve seen tools convert the other way around like onnx muellerzr/fastinference/blob/master/fastinference/ and mmdnn. I understand this as run research work using pytorch/fastai first, then convert them to deployment-ready models.

So what I am looking for is something like tf-fit but this is not updated for fastai2. Does anyone know any other tools like this? And, why is it so rare to find these tools?

Hope this makes sense.

Cheers, Hud