Tracking by Self-detection

This is the repository containing all the implementations of Tracking by self-detection.

Video Demo

Results:

1. Our model : SSL + EfficientDetD0 + Deepsort with trained association model

The ReID model was trained for a total of 54 epochs and had an accuraccy of 92.94% on the test data.

For running the model on test videos:

1. Modify the dataset path in deepsort.py
2. Run

python deepsort.py

For evaluation of results:

1. clone the github repository git clone https://github.com/JonathonLuiten/TrackEval
2. change directory to :

cd TrackEval\scripts

3. run the script :

python .\run_mot_challenge.py 
    --GT_FOLDER ".dataset\gt" 
    --BENCHMARK "MOT16" 
    --TRACKERS_FOLDER ".\dataset\trackers" 
    --SPLIT_TO_EVAL "test"

Results

• Detailed results of our method is given in detailed_results.txt

Method	HOTA(%)	MOTA(%)	IDF1(%)	IDSW	FRAG
SORT	30.71	75.7	53.40	11684	216
DEEPSORT	64.15	84.17	74.04	491	431
BYTETRACK	50.27	36.89	54.41	71	270
Proposed	67.86	84.74	80.94	165	335