How to install

Setup

make venv
. ./venv/bin/activate
make develop

Commands

Ground truth visualisation - visualise ground truth for test dataset**

deep-sort ground-truth

You as a result you should see this:

Tracker runs the DeepSORT algorithm on all datasets one by one

deep-sort run -fe [features extractor name] -d [detections provider name]

Note: Check out deep-sort run -h for a detailed overview of available options

After each video processing you can find result metrics in terminal:

Results

The main evaluation stages are published in this Colab.

Best results:

yolov10x, osnet_ain_x0_75

                    |F1        |Recall    |Precision |
KITTI-17            |   0.83291|   0.72834|   0.97254|
MOT16-09            |   0.72158|   0.56756|   0.99035|
MOT16-11            |   0.74241|   0.59331|    0.9916|
PETS09-S2L1         |   0.68958|   0.67554|   0.70422|
TUD-Campus          |   0.70491|   0.60563|   0.84313|
TUD-Stadtmitte      |    0.8336|   0.74521|   0.94591|