LemonGraZ
LemonGraZ
@felipetobars Regarding the fact that Y11-P6 has less GFLOPs and performs worse might have to do with the fewer number of channels in the P2 layer, like I mentioned above....
According to ChatGPT and Gemini the following points are related to that decision. To me that sounds pretty reasobable. - **Quantization of Regression Outputs:** Instead of directly predicting four offsets...
@Y-T-G Thanks, under what key-word do I find that? Is this part of the Distributed Focal Loss (DFL) and thus more related to that paper?