Yunsheng Li

Results 72 comments of Yunsheng Li

step (1) -> (4) can be iterated. You can just ignore from fcn8s_LSD import FCN8s_LSD, it is only for VGG net.

I'm not sure. The gap seems to be large. In fact micronet is developed mainly for CPU usage. The speed should not be very fast due to the existence of...

I'm sorry. I'm not able to solve out of memory problem. Maybe your GPU doesn't have enough memory. But I think 12Gb is enough.

That's very weird to me. Up till now, no one has even reported the out of memory problem when testing the model. 12Gb is very enough. Do you resize the...

The size of the input images have been set in the file of BDL/data/__init__.py. At least, you should run the evaluation.py code and check the image size you load.

I'm not sure. I use pytorch 0.4.0. But I don't think it is hard to transfer the code for pytorch 1.7.

The keypoint detection code is written by one of the coauthors, you may need to ask him directly [email protected]

DepthSpatialSepConv is the Micro-Factorized Depthwise Convolution. ChannelShuffle and ChannelShuffle2 are the same and I use different names for ablation only. I should have removed the ChannelShuffle2. For Micro-Factorized Pointwise Convolution,...

Thanks for letting me know the warning. I think it does not influence the implementation of the other codes. Since I cannot produce the warning on my side, could you...

I think it is probably caused by the mismatch of Pytorch version since I cannot reproduce this warning on my side. Thanks for letting me know.