daquexian

@i-amgeek Thanks for your report! Generally reshape is hard to implement since the difference between NCHW (ONNX) and NHWC (dabnn). However, the cases that reshape is before gemm/matmul are trivial....

Thanks for your issue! You are right, arm neon only supports xor. The coefficients in the following batchnorm layer are modified to make the result correct. This step is done...

I will update the document for this information soon.

> Hi, > > I think the error is because I have 0 on dim_value. I edit my onnx model and regenerate new onnx. Now I have another error: Unsupported...

Hi :) I'm also a user (and a [contributor](https://github.com/Tencent/ncnn/graphs/contributors)) of ncnn. I agree that a fork of ncnn with accelerated binary convolution implementation is exciting. However, I'm afraid I have...

Could you please provide your onnx model? Thanks!

Thanks for your attention! You might want to check out https://github.com/JDAI-CV/dabnn-example . I'll also add some detailed document soon. On Thu, Oct 17, 2019, 10:56 PM Bil17t wrote: > First,...

Which do you want to do, benchmark some models on your side, or do actual inference? Thanks!

For verifying the accuracy, I think you can modify the code based on https://github.com/JDAI-CV/dabnn-example and check the output currently.

And if you want to verify the accuracy more carefully, you can check out 'binaries/run.cpp' about how to send a given input tensor to the model and retrieve the output.