To preserve the low-level image features of the image, we replace the shallow stem with a deep stem composed of multiple 3 × 3 convolutional layers following Res2Net [14].
But in your code i don't see any Res2Net-style bottleneck modifications. Did you use it?
In p. 4.2.1 you wrote:
But in your code i don't see any Res2Net-style bottleneck modifications. Did you use it?