Why is there no downsampling in the first convolution layer of resnet for imagenet dataset?

[linlany]

https://github.com/G-U-N/PyCIL/blob/07c9114d8b9be1534185ab1d9b36f1709867c7f0/convs/resnet.py#L174 This setting is usually applied to low-resolution datasets such as cifar100 to avoid too small feature maps. If the imagenet dataset also uses this setting, it will result in feature maps that are four times larger than the original resnet, which consumes a lot of memory and is not a common setting.