Closed YohayAiTe closed 1 year ago
Looks really cool! Apart from the one minor comment, would you mind adding a unit test for the new behavior to onnxruntime_test.go
? I'll be happy to merge it after that. I may use some of what you added to implement the features requested in #22 (even though the requester no longer seems interested in that one).
I added a test and fixed the doc comment.
Looks great, thank you for working on it!
Realized I had forgotten to tag this. This feature is now included in tagged version v1.4.0
I added an option to pass a null tensor to
DynamicAdvancedSession.Run()
, which will allocate the tensor and populate the slice with the new tensor. This is similar to the behaviour of the C/C++ API, and can be useful when the output shape can't be predicted at ahead of time (such as image object detection). Example usage: