Three files train2017.txt, val2017.txt and test2017.txt contain face labels for COCO 2017 dataset.
Each file contain lines in format "file x1 y1 x2 y2", where file is name of image in the specific COCO part, (x1, y1) upper-left corner of face rectangle, (x2, y2) - lower-right corner. All coordinates are presented as floating point numbers in range [0, 1], relative to width and height of the specific image.
All images are labeled with RetinaFace detector (RetinaFace-R50 currently).