coco2mot

Purpose?

This module has utilities to transform a COCO scheme into MOT format.

Dependencies?

• pandas
• numpy

Features?

Two types of features:

• module features
• command-line features

module features?

coco2mot(fname)

from coco2mot import coco2mot

fname = r"coco/path.json" # file name
df = coco2mot(fname)
df.columns # [frame, id, tl_x, tl_y, width, height, conf, x,y,z]

given a coco json path, it returns a dataframe in det.txt mot format

mot_backbone(fnames)

from coco2mot import mot_backbone

fnames = ["vid1", "vid2", "vid3"] # file names
opath = r"/output/path" # output path
mot_backbone(fnames, opath)

Output:

└─ opath/
    ├── vid1
    │   ├── det
    │   ├── gt
    │   ├── img1
    ├── vid2
    │   ├── . . .
    ├── . . .

give a list of names, it makes a sequence directory for each name in list. All these directories are empty.

command-line features?

mot_backbone.py

python ./cmd/mot_backbone.py --pname parent_dir --opath output/dir --cpath /coco/files/dir

Output:

└─ output/dir/parent_dir
    ├── coco_json_1
    │   ├── det
    │   ├── gt
    │   ├── img1
    ├── coco_json_2
    │   ├── . . .
    ├── . . .

coco2mot.py

python ./cmd/coco2mot.py --mpath /mot/dir/ --cpath /coco/files/dir

Output:

└─ output/dir/pname
    ├── coco_json_1
    │   ├── det
    │   │   ├── det.txt
    │   ├── gt
    │   ├── img1
    ├── coco_json_2
    │   ├── det
    │   │   ├── det.txt
    │   ├── gt
    │   ├── img1
    ├── . . .

Adds a det.txt file in mot format from coco json files specified in --cpath.

vid2frames.py

python ./cmd/vid2frames.py --mpath /mot/dir --vpath /videos/dir --meta vid/to/dir/map.csv

Output:

└─ output/dir/pname
    ├── coco_json_1
    │   ├── det
    │   │   ├── det.txt
    │   ├── gt
    │   ├── img1
    │   │   ├── 000001.jpg
    │   │   ├── 000002.jpg
    │   │   ├── . . .
    ├── coco_json_2
    │   ├── det
    │   │   ├── det.txt
    │   ├── gt
    │   ├── img1
    │   │   ├── 000001.jpg
    │   │   ├── 000002.jpg
    │   │   ├── . . .
    ├── . . .

Adds frames %06d.jpg that correspond to each specified mapping in map.csv.

This csv contains two columns with any names where the first column represents the video names stored in --vpath and the second column represents the corresponding coco json file that it corresponds to. For example, our above example could be seen as:

vid_name,dir_name
vid1,coco_json_1
vid2,coco_json_2
. . .

NOTE: as you can see from above, mot_backbone.py, coco2mot.py, and vid2frames.py all sequentially build around each other.

New Features?

A few simple features are offered in this repository. Any new functionalities that you create from the extension of these modules are welcome!