Skip to content

Latest commit

 

History

History
83 lines (75 loc) · 2.09 KB

README.md

File metadata and controls

83 lines (75 loc) · 2.09 KB
Raw

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
  1. run the script :
python .\run_mot_challenge.py 
    --GT_FOLDER ".dataset\gt" 
    --BENCHMARK "MOT16" 
    --TRACKERS_FOLDER ".\dataset\trackers" 
    --SPLIT_TO_EVAL "test"

Results

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