masahi edited a comment on pull request #6472: URL: https://github.com/apache/incubator-tvm/pull/6472#issuecomment-692943139
@yongwww to be clear, this translation works for box_coder from maskrcnn-benchmark you mentioned above? https://github.com/facebookresearch/maskrcnn-benchmark/blob/4ac0a3b440320ff0d0159a280f2897d1f9f996f3/maskrcnn_benchmark/modeling/box_coder.py#L85-L93 That looks similar to the test case @t-vi provided. By "work", of course I mean we get the same output as pytorch would. > I would suggest adding a warning and todo in this mapping since some other merged in-place op mapping should has the similar issues, don't have a quick solution for this in my mind at this point. Any suggestion? I'm ok with this, if we can somehow detect inplace op patterns that we cannot support, we should immediately abort, rather than giving a warning and emitting a wrong relay graph. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected]
