You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
HRNets are designed as the backbone for many vision tasks.
If you want to reproduce the best performance in Cityscapes leaderboard, you also need OCR. OCR is another work, and may be released after the paper accepted.
Thanks for your attention.
Which part uses an OCR?
And how much would be the performance without OCR?
Can we use the published pre-trained models without this OCR, ie to do some fine-tuning on the Cityscapes?
I found the best performance in Cityscapes leaderboard(83.7). How can I reproduce this result?
The text was updated successfully, but these errors were encountered: