Comment on ONNX runtime: Cross-platform accelerated machine learningComments−ThouYS3yonnx is nice in principle, but pretty limited. core ops like non-max-suppression can't get properly converted. also model deployment is not great, memory consumption and control thereof worse than with tensorflow.
Comments
onnx is nice in principle, but pretty limited. core ops like non-max-suppression can't get properly converted. also model deployment is not great, memory consumption and control thereof worse than with tensorflow.