Skip to content

Latest commit

 

History

History
126 lines (89 loc) · 3.45 KB

README.md

File metadata and controls

126 lines (89 loc) · 3.45 KB
Raw

LITE

LITE: A Paradigm Shift in Multi-Object Tracking with Efficient ReID Feature Integration

Jumabek Alikhanov, Dilshod Obidov, Hakil Kim

arXiv 2409.04187

Abstract

The Lightweight Integrated Tracking-Feature Extraction (LITE) paradigm is introduced as a novel multi-object tracking (MOT) approach. It enhances ReID-based trackers by eliminating inference, pre-processing, post-processing, and ReID model training costs. LITE uses real-time appearance features without compromising speed. By integrating appearance feature extraction directly into the tracking pipeline using standard CNN-based detectors such as YOLOv8m, LITE demonstrates significant performance improvements. The simplest implementation of LITE on top of classic DeepSORT achieves a HOTA score of 43.03% at 28.3 FPS on the MOT17 benchmark, making it twice as fast as DeepSORT on MOT17 and four times faster on the more crowded MOT20 dataset, while maintaining similar accuracy. Additionally, a new evaluation framework for tracking-by-detection approaches reveals that conventional trackers like DeepSORT remain competitive with modern state-of-the-art trackers when evaluated under fair conditions.

Efficient ReID feature extraction via the LITE paradigm

Environment

We use Python 3.10.12

python3.10 -m venv myenv
source myenv/bin/activate
pip install -r requirements.txt

ultralytics

git clone https://github.com/humblebeeintel/ultralytics.git

Demo

python lite_deepsort_demo.py --source demo/VIRAT_S_010204_07_000942_000989.mp4

Experiments Settings

Datasets

Download our prepared datasets and put them under LITE/datasets in the following structure:

datasets
   |———MOT
   |     └———train
   |     └———test
   └———PersonPath22
   |     └———test
   └———VIRAT-S
   |     └———train
   └———KITTI
         └———train
         └———test

Checkpoints

Download checkpoints and put them under LITE/checkpoints:

checkpoints
└── FastReID
    ├── bagtricks_S50.yml
    ├── Base-bagtricks.yml
    ├── deepsort
    │   ├── ckpt.t7
    │   └── original_ckpt.t7
    └── DukeMTMC_BoT-S50.pth

FastReID

bash scripts/setup_fastreid.sh

yolo_tracking

git clone https://github.com/humblebeeintel/yolo_tracking.git

Running Experiments

Use the following command to run experiments with different datasets and splits:

bash scripts/run_experiment.sh -d <DATASET> -s <SPLIT>

Running FPS Experiments

Use the following command to run fps experiment with specific sequence from datasets:

bash scripts/run_fps_experiment.sh -d <DATASET> -s <SPLIT> -q <SEQUENCE>

Solutions demo with LITEDeepSORT

bash demo/download_solutions_demo_videos.sh

Object Counter & Heatmap

python lite_deepsort_solutions_demo.py \
--source videos/shortened_enterance.mp4 \
--solution object_counter
           heatmap

Parking Management

python lite_deepsort_solutions_demo.py \
--source videos/parking.mp4 \
--solution parking_management

Multi Object Tracking made easy and accessible

Code is coming soon!