cangcz/AnchorCrafter-finutune

Dataset Structure

The video_cut directory contains the original videos. Each file name follows the format:
<person_id>_<object_id>.mp4, where the first number represents the person ID, and the second number represents the object ID.

For example:

• video_cut/1_0.mp4 corresponds to the person video people_cut/1.mp4.
• The associated object images are masked_object_cut/0_0.jpg, masked_object_cut/0_1.jpg, and masked_object_cut/0_2.jpg.

Additionally, the dataset includes annotations:

• Human Pose Data: Located in the video_pose directory.
• Hand Pose Data: Stored in hand_cut.
• Object Depth Information: Found in depth_cut.

All file names in these directories remain consistent with the original video filenames to ensure seamless cross-referencing.