Regarding the ground truths for MultiviewX

[luorix1]

trafficstars

Hello, I recently managed to make a naive, rough Unity project that produces data similar to that of MultiviewX.

I am hoping to utilize your python files to create annotations and camera calibrations so I can try this out on MVDet. For your 2D, 3D bounding boxes, could you explain the format in which you saved your bounding boxes? I am having some trouble understanding the Camera~~.txt files in the matchings/ directory.

Thank you, Jinwoo Hwang

[hou-yz]

Hi,

Thank you for your question. In essence, the camera~.txt files provide 3d bounding boxes in both 3d coordinates and their 2d correspondences, both generated from unity. Sorry, but I cannot recall the exact meaning of the coordinates, and you might need a little exploration yourself.

Best, Yunzhong

[luorix1]

Just in case anyone needs this for future reference, each row seems to consist of the following.

• first column: frame number (0~399)
• second column: person ID (PID)
• third column onward: 3 (or in case of 2D bounding boxes, 2) coordinates for each vertex of the cuboid 3D bounding box and the feet of a person.