Open swarajnanda2021 opened 8 months ago
hi! just curious that did you run into this issue? how did you solved it?
I was implementing CoATNet myself and sought this repo for inspiration. It did not work, so while debugging I had to re-read the paper several times. Finally I understood the problems and accordingly found a solution. Of course, GPT4 helped a lot here.
how was the training results of the model? did you use downsampling ? which one gives better results? with downsamling or without downsampling?
I've found the MBConv to have some computational inconsistencies. The following corrected code works, where I've changed the stride of the projection operation (
self.proj
) and moved it out of theif downsample
statement. Further, the squeeze and excite block has been appropriately initialized (I've added my squeeze and excite block too here for completeness). I've also added the channel projection operation on thedownsample is false
branch of MBConv forward method: