-
Notifications
You must be signed in to change notification settings - Fork 187
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
What is the current accuracy of the uploaded project and trained model respectively? #13
Comments
The mAP of trained model is around 57.5,and this repo can get 60 or so. @ZhangGongjie |
May I ask how did you perform data augmentation to keep data balance? Please kindly advise. (・ω´・ ) |
Actually, I did not do any data balance. @ZhangGongjie |
That's amazing...... Actually, I am currently working on this dataset. I did not perform data balancing and get extremely low accuracy on some object instances like 'ground-track-field' and 'helicopter'. Besides, could you please kindly advise how did you fuse the extremely large objects like ground-track-fields and roundabouts? Usually these kind of objects are split into parts when performing cropping to DOTA dataset.(・ω´・ ) |
You can refer to my data process code. here. @ZhangGongjie |
The size of crop image is 800*800 with overlap 200 both in height and width. @ZhangGongjie |
From my experience, even an overlap of 200 cannot guarantee such large object fall into a single image without splitting. |
Yes, It's hard to guarantee coverage all objects. |
非常感谢!共同进步! :) |
As above.
Thank you!
The text was updated successfully, but these errors were encountered: