John Brandt

Adding onto this, it doesn't seem that the # params matches the paper. Torch tells me that `yolov9-c` has `51043480` parameters and `yolov9-e` has `69437144` parameters -- Nevermind, I see...

Hi - Sorry for the late response! This work was published in the journal of computational social science, here: https://link.springer.com/article/10.1007/s42001-019-00061-9 The paper talks about the collection and labeling of data,...

A few thoughts -- The original layer norm paper advise against applying it to CNNs: "With fully connected layers, all the hidden units in a layer tend to make similar...

It seems to me that layer norm helps stabilize training but hurts the final test accuracy. In the layer norm paper, the authors warn against using it in CNNs because...

Got this to work, I refactored the code to work with MMDet 3.X