I only have a RTX 4090 GPU, and I only can set the total batch_size to 8. However, no matter what learning rate I choose, I can not train the model to reproduce the 85.91 mAP(I only get 85.5 mAP or even worse). Is there any methods to further increase
the performance? And I found that training process is not stable(the best result may be the 75 repoch not the last epoch, and the best result varies among the different training processes(e.g. best mAP between 85.2 and 85.5 for the same hyper-parameters)). Is your training process the same with mine? Hoping for your reply!
I only have a RTX 4090 GPU, and I only can set the total batch_size to 8. However, no matter what learning rate I choose, I can not train the model to reproduce the 85.91 mAP(I only get 85.5 mAP or even worse). Is there any methods to further increase the performance? And I found that training process is not stable(the best result may be the 75 repoch not the last epoch, and the best result varies among the different training processes(e.g. best mAP between 85.2 and 85.5 for the same hyper-parameters)). Is your training process the same with mine? Hoping for your reply!