Tian, Feng comments

Results 52 comments of


                                            Tian, Feng

__builtin_shuffle not supported in icc

same problem. nobody complains it?:(

BUG: Unexpected exception TypeError("_post_eval_hook() got an unexpected keyword argument 'accuracy'") happened during tuning.

thanks for the feedback. we will take a look on that

Quantization Not Fusing Pad to Conv2D

thanks for raising such an issue. INC supports Pad + conv fusion. The problem is if the pad op’s tensor value is not all 0, it may trigger TensorFlow int8...

Quantization AssertionError "inputs len must equal with input_tensor"

could you pls paste the whole log? As you specified the metric and eval_dataloader, neural compressor should construct the evaluation process in internal. the duration should not be 0. but...

Use TorchVision's Model Registration API to fetch available backbones

@datumbox many thanks for raising this. we will have a check on that and get back to you asap

Noise in NMS threshold filtering (GetMaxScoreIndex)

thanks for the reminder, we will merge this update to next release.

Noise in NMS threshold filtering (GetMaxScoreIndex)

@MinhazPalasara @matt-ny I just reviewed the fix. I have a question about it: the default pair values in the vector are all 0. why using .at() has problem? for example:...

Noise in NMS threshold filtering (GetMaxScoreIndex)

Thanks for your elaboration, the idx 0 is used wrongly with original code. I will merge your fix to our repo.

Noise in NMS threshold filtering (GetMaxScoreIndex)

@matt-ny this fix would make unit test fail. you can reproduce it by "cmake .. && make test.testbin && ./test/test.testbin --gtest_filter=\*NMS\*".

Noise in NMS threshold filtering (GetMaxScoreIndex)

I think I have known what happened after debugging. we combined vector push_back() with openmp parallel for primitive without protect, it will bring issue at line 2266.(or wrong value or...