search

dendorferpatrick / MOTChallengeEvalKit

263 stars 42 forks source link

MOT20\train\MOT20-01\gt\gt.txt data format not consistent #4

Open WurmD opened 4 years ago

WurmD commented 4 years ago

https://github.com/dendorferpatrick/MOTChallengeEvalKit/blob/master/MOT/README.md states

Data Format: The file format should be the same as the ground truth file, which is a CSV text-file containing one object instance per line. Each >line must contain 10 values: <frame>, <id>, <bb_left>, <bb_top>, <bb_width>, <bb_height>, <conf>, <x>, <y>, <z>

The data at https://motchallenge.net/data/MOT20/ in MOT20\train\MOT20-01\gt\gt.txt does not contain 10 values, only contains 9 in each line

abhishekchopde commented 3 years ago

@WurmD did you find an answer to this issue? I am also wondering about the same thing. What do the 9 columns represent exactlt? I am not sure about the last 3 fields.

AbdurahmaanYuusuf commented 3 years ago

it is explained in this paper: MOT16: A Benchmark for Multi-Object Tracking

DerekGloudemans commented 3 years ago

To summarize the paper, the columns are as follows: <frame>, <id>, <bb_left>, <bb_top>, <bb_width>, <bb_height>, <conf>, <class>, <visibility>

where visibility is 1 if the object is completely visible and 0 if the object is completely occluded, and classes are: Pedestrian 1 Person on vehicle 2 Car 3 Bicycle 4 Motorbike 5 Non motorized vehicle 6 Static person 7 Distractor 8 Occluder 9 Occluder on the ground 10 Occluder full 11 Reflection 12

Only ground-truth pedestrians are included in scoring

JAYCHOU2020 commented 3 years ago

scoring

, I would like to ask you,when i make my own data annotations , how is the visibility rate mentioned in the annotation obtained?

DerekGloudemans commented 3 years ago

Not 100% sure, try the paper. I'm guessing they combine truncation (how much of the object is within the frame) and intersection with other objects, possibly taking into account which objects are in front of one another.