Why does HybridNets require larger GPU memory?

[yaoshanliang]

In YOLOP, I can set batch_size to 64 in a 3090 sever. However, in HybridNets, the batch_size can only be set to 8. Otherwise, the program shows the message "Out of Memory".

I have made some modifications according to the readme file and some issues in this project, including python train.py --cal_map False, python train.py --conf_thres 0.5, but it still trains slowly.
Do I need to modify some other configs in your project?

[wxzb625]

return F.conv2d(input, weight, bias, self.stride,
RuntimeError: Given groups=1, weight of size [64, 32, 3, 3], expected input[2, 24, 98, 162] to have 32 channels, but got 24 channels instead
你好，看到您对这个项目了解很深，对于这个项目，当我把compound_coef=3改为compound_coef=1时，也就是把网络由b3替换成了b1，会出现上述的错误，为什么会出现层数不匹配的错误，希望得到您的解答。