Skip to content

Latest commit

 

History

History
12 lines (9 loc) · 887 Bytes

README.md

File metadata and controls

12 lines (9 loc) · 887 Bytes
Raw

CrowdHuman Dataset by MEGVII

  • Official web site: https://www.crowdhuman.org/

  • Reference:

  • When converting CrowdHuman annotations to YOLO txt files,

    • I discard all "mask" objects. The "mask" objects in the CrowdHuman dataset are not real humans. They are usually reflections of humans, or pictures of humans in billboards or advertisement posters.
    • I use "hbox" (head) and "fbox" (full body) annotations of all "person" objects. Note that the "fbox" annotation might include body parts which are "ocluded" in the scene.
    • In the final YOLO txt files, there are 2 classes of objects. Class 0 is "head", and class 1 "person".