ADAM
a solution might be 1. using average pooling so regardless of the output size of the image, it will shrink it down to a scaler of shape:[1] 2. just use...
I'm not sure what you trying to so, hope you weren't thinking about training the model from scratch. 4 gpus are more than enough and you can use transfer learning...
yeah, that is mainly because pytorch expects a shape of [N,C,H,W] even as of today (8/9/2025).
7*7 is wrong, and it will throw an error when you start training. just posted an issue with the full explanation.