YuzhouPeng

@lukemelas

The efficientnet link is https://github.com/qubvel/efficientnet

same error, have anyone solve it?

I tried to use 3090 to test image, and the average inference speed is 160ms per image(1918x1280, batch size1), and maybe some information in CMakeList (sm version )influence the inference...

I can not test 1080ti performance but I searched and find a similar issue: https://github.com/NVIDIA/TensorRT/issues/1221 maybe use different cudnn can help

yes, if you trained with grayscale image, you need to change this to 1 to fit the channel number, meanwhile you should also change the channel number of input in...

Yes, you need to compile code to a library (.so file) and use pybind11 to call the function. Maybe I will add plugin in the future if I have time...

你有使用链接里的pytorch unet训练权重吗？我看了下好像weights翻倍了

如果训练按照步骤来的话，可能是factor的问题，我抽时间fix下把

you can try to convert gray image to three channel, and train them.