Exactly what happened to me. Let me guess, yolo11s? I think I had the same issue.

11m. But I got similar scores with 11s and 11l.

Have you tried augment = True, lower confidence and maybe tuning IOU to like 0.55 for inference

Yes, I have. But in an ensemble. Trying to tune individual models now

Wow, I just remembered I haven't tried ensembling 😅. Thanks @nymfree

you're in a good position. you're about to hit 0.82x!

I have seen you've pushed massively. Nice work done @nymfree. I tried the ensembles on weaker models and I think it is quite good.

Thanks. Tuning parameters seem to be very very important. Even when ensembling. Now knocking at the door of club 800.

Yes I've seen that. Congrats big man🔥🔥🔥

Indeed hyperparameter tuning is indeed very necessary. Sigh😮‍💨. I only did few at first. Thanks @nymfree 😊

Thanks. Good news is we got 3 days and ~40 subs to experiment with. Hopefully, we are not overfitting to the public LB and the shake-up will be favorable to us.

Hopefully 😅.