-
Notifications
You must be signed in to change notification settings - Fork 115
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Cannot get the AP result in the paper #37
Comments
Could you provide more details about how you modified the settings? It will be great if you can provide your training log file. |
Here is the modified settings: SOLVER: Other parts are not changed. |
Try:
|
Thanks for your reply. I tried this warmup settings and only got an AP of 35.8365. Here is part of my training log ( I have to delete some lines since it exceeds the maximum character length of the comment) [08/06 16:42:57] detectron2 INFO: Rank of current process: 0. World size: 1 sys.platform linux PyTorch built with:
[08/06 16:42:59] detectron2 INFO: Full config saved to output/yolof/R_50_C5_1x/config.yaml [08/06 16:43:39] d2.engine.train_loop INFO: Starting training from iteration 0 [08/07 16:27:45] d2.engine.defaults INFO: Evaluation results for coco_val_rgb in csv format: The settings different from the orginal settings are below: |
I checked the setting once again. The steps should be modified. The initial steps Try:
You could get reasonable results this time. |
It works! Thank you for your help! |
Hello! I'm trying to re-train this YOLOF in a single 3090 GPU with the batchsize of 32. However, even though I tried to modify some of the parameters like base-lr, steps and something else, the best AP I can get is 35.7688, which is much worse than the AP of 37.7 in your paper. Do you have any suggestions about how to modify base-learning rate, steps and warmup settings?
Thank you!
The text was updated successfully, but these errors were encountered: