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

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

Jumabek Alikhanov, Dilshod Obidov, Hakil Kim

arXiv 2409.04187

Overview

LITE (Lightweight Integrated Tracking-Feature Extraction) introduces a groundbreaking approach to enhance ReID-based Multi-Object Tracking (MOT) systems. By integrating appearance feature extraction directly into the detection pipeline, LITE significantly improves computational efficiency while maintaining robust performance. Utilizing CNN-based object detectors like YOLOv8 and YOLO11, LITE enables real-time tracking, making it ideal for resource-constrained environments.

---

Key Features

• Efficient Integration: Combines appearance feature extraction within the detection process.
• Lightweight Design: Tailored for real-time applications on resource-limited devices.
• Performance Optimization: Demonstrates notable FPS improvements across multiple trackers while retaining competitive accuracy.

---

Experimental Results

We evaluated LITE using YOLOv8m with the following settings:

• Confidence Threshold: 0.25
• Input Resolution: 1280

Tracker	MOT17 HOTA ↑	MOT17 FPS ↑	MOT20 HOTA ↑	MOT20 FPS ↑
DeepSORT	43.7	10.5	24.4	8.5
StrongSORT	44.5	4.5	26.1	2.6
Deep OC-SORT	43.7	10.3	24.9	8.9
BoTSORT	40.8	10.6	21.1	9.4
LITE:DeepSORT	43.0	26.7	25.2	15.9
LITE:StrongSORT	42.4	29.7	25.2	22.9
LITE:Deep OC-SORT	43.4	34.8	25.4	19.6
LITE:BoTSORT	40.8	38.2	21.1	31.8

---

Installation

Follow these steps to set up the LITE tracking system:

1. Clone the Repository

# Clone the LITE Tracker Repository
git clone https://github.com/Jumabek/LITE.git
cd LITE

2. Set Up Python Environment

# Create and activate a virtual environment
python3.10 -m venv myenv
source myenv/bin/activate

# Install dependencies
pip install -r requirements.txt

3. Clone Additional Dependencies

# Clone supplementary repositories
git clone https://github.com/Jumabek/ultralytics.git
git clone https://github.com/humblebeeintel/yolo_tracking.git
git clone https://github.com/humblebeeintel/TrackEval

4. Set Up FastReID

bash scripts/setup_fastreid.sh

---

Dataset Preparation

Download the prepared datasets from this link and organize them as follows:

LITE/datasets
   |-- MOT
   |   |-- train
   |   |-- test
   |-- PersonPath22
   |   |-- test
   |-- VIRAT-S
   |   |-- train
   |-- KITTI
       |-- train
       |-- test

---

Checkpoints

Download the required checkpoints from this link and place them under LITE/checkpoints:

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

---

Running Experiments

1. Run Tracking and ReID Experiments

bash scripts/run_experiment.sh -d <DATASET> -s <SPLIT> -t <TRACKER> -m <YOLO_MODEL>
# TRACKER options: "SORT", "LITEDeepSORT", "DeepSORT", "StrongSORT", "LITEStrongSORT", "OCSORT", "Bytetrack", "DeepOCSORT", "LITEDeepOCSORT", "BoTSORT", "LITEBoTSORT"

# YOLO_MODEL options: all models of YOLO from yolov8 to yolo11

2. Running the ReID Evaluator

python reid.py --dataset <DATASET> --seq_name <SEQ_NAME>  --split <SPLIT>  --tracker <ReID_MODEL> --save

Demo

Download demo videos

bash demo/download_solutions_demo_videos.sh

Basic Tracking Demo

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

Object Counter & Heatmap

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

Parking Management

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

---

Citation

If you use LITE in your research, please cite our work:

@misc{alikhanov2024liteparadigmshiftmultiobject,
      title={LITE: A Paradigm Shift in Multi-Object Tracking with Efficient ReID Feature Integration}, 
      author={Jumabek Alikhanov and Dilshod Obidov and Hakil Kim},
      year={2024},
      eprint={2409.04187},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2409.04187}, 
}