PaddleDetection Make LCPAN also pass the activator to DepthwiseSeparable

Without this, nn.Hardswish() will be selected in ConvBNLayer, which causes the opset version to be bumped to 14, in turn, and in my case, breaks ESP-DL ONNX quantization, which only supports 13.

I would also suspect that this is more correct also in other cases as one would not be likely to not want to use a specific activation function everywhere, would one supply it.

Also, it aligns with how the act-parameter is handled in other cases.

(accidentally first did the PR against a release branch)

Jan 13 '25 21:01 nicklasb

Thanks for your contribution!

Jan 13 '25 21:01 paddle-bot[bot]

All committers have signed the CLA.

Jan 13 '25 21:01 CLAassistant

@LokeZhou @jzhang533 This causes issues and probably make the output break several use cases, please include this change.

Mar 02 '25 13:03 nicklasb

@LokeZhou @jzhang533 isn't this an obvious bug? I have to correct this each time i use a new version of PaddleDetection?

May 10 '25 19:05 nicklasb

@LokeZhou @jzhang533 isn't this an obvious bug? I have to correct this each time i use a new version of PaddleDetection?

I agree

May 12 '25 08:05 simoberny

@zhangyubo0722 , @SigureMo , can you please review this? It is very annoying to have to patch this every new version.

@simoberny Sorry, accidentally requested a review from you.

May 20 '25 22:05 nicklasb