Question about retinaface_gt_v1.1 dataset


The image above shows the visualization of 0_Parade_marchingband_1_657.jpg

bboxes and landmarks are from retinaface_gt_v1.1 dataset

As can be seen from the training code, landmark score 1 means invisible, 0 means visible.
landmark color: black indicates visible point, red indicates invisible point.

But as can be seen from the above picture, these points are all visible.

Start by checking the retinaface_gt_v1.1 annotation for 0_Parade_marchingband_1_657.jpg and the training code's landmark-score convention described in the issue. Done means determining whether the visibility labels or their visualization are incorrect and documenting the explanation or required correction.

Written by the indexing model from the issue text.