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
Thanks for confirming. I didn't know that TTA is about inferencing image at different scale, found this great tutorial from YOLOV5 repo.
But, I think there is some bug that does not let proper scaling of the keypoints/landmarks coordinates when using TTA with/without image scaling by specifying the arguments --augment --img 832 or augment. Some of the landmarks coordinates are not being scaled as per the image size and end up plotted at random places in the frame.
Weight file url of
yolov7-w6
andyolov7-w6+TTA
refers to the same Google Drive URL inREADME.md
The text was updated successfully, but these errors were encountered: