Martin Han

Same problem... If I ask others to annotate the file, then the path is not the same, and I cannot open it on my machine. I would prefer vott to...

DWConv2D with bias is resolved, but there is something wrong with deconv. I am working with ncnn guys to resolve it.

Can you provide the model for debugging?

Thanks! Can you also provide the original keras model comparing?

Here is the keras2ncnn debug mode output (by inputing random data), it seems like the network forward path is completely identical, so the problem may be in the pre-processing stage...

Something like this happened before... but I just not able to find where the accuracy mismatched.

Maybe you can also give me a test picture and expected output

Cool, let me have a look.

> @MarsTechHAN > > How did you print [these](https://github.com/MarsTechHAN/keras2ncnn/issues/29#issuecomment-815892835) values ? > > If possible can you please help me with the program which prints it ? It's keras2ncnn's build-in...

> @MarsTechHAN > > One thing I noticed is this, in keras the dimensions of input are `type: float32[?,128,128,3]` and in NCNN is `type: float32[128,128,3]` > > Keras is 4D...