[Question] Your take on how to make the classification head better

[HarshitSheoran]

Hi @ShihuaHuang95

Thank you for this amazing work, I have been exploring this for about a week and I get more impressed the more I look into it

For my task, the detection of the object is an extremely easy task, they all look somewhat similar so making a correct box is easy for the model but to differentiate them into 23 classes is much harder

How would you advice, to look into such a task, what parts of the pipeline or model should I be looking to modify/experiment with to potentially get better performance in classification

Kind Regards