Question about "instance_id" in the json file of MOT task

[LilyDaytoy]

Hi, sorry I' m a beginner~ I' d like to ask that is instance_id in json annotations file of MOT task a "global id" to all videos in the dataset, or it is numbered within a video? like: vid1: instance 1,2,3 vid2: instance 4,5... or vid1: instance 1,2,3 vid2: instance 1,2... Is there any convention followed by the video object detection community?

[dyhBUPT]

Hi, the instance_id is numbered with in the video. More details can be seen in: https://github.com/open-mmlab/mmtracking/blob/master/tools/convert_datasets/mot/mot2coco.py

Best wishes.

[LilyDaytoy]

Hi~ but when I tried the json generate in demo, why for image_id >=601 (1-600 belongs to the first video, starts from 601 is the second video), there is no instance_id < 62

[dyhBUPT]

Oh, sorry for my error. The exact statement is:

• For the raw annotations (in .txt) of MOTChallenge, it's numbered within the video.
• For the generated annotations (in .json), it's the global id across all videos.

in details: https://github.com/open-mmlab/mmtracking/blob/be8a7afbd719b7846a76caca74dd7f331036cdc3/tools/convert_datasets/mot/mot2coco.py#L198-L203

[LilyDaytoy]

ohh, get it, thanks a lot :D