Readme

Vehicle Re-Identification Framework

This repository contains the code and resources for training, testing, and evaluating vehicle re-identification models using deep learning techniques. The directory structure and main components of this framework are detailed below.

---

Directory Structure

.
├── .github/
├── docs/
├── logs/
│   └── resnet50-veri/
│       └── .gitkeep
├── src/
│   ├── losses/
│   │   ├── __init__.py
│   │   ├── center_loss.py
│   │   ├── cross_entropy_loss.py
│   │   ├── hard_mine_triplet_loss.py
│   ├── models/
│   │   ├── __init__.py
│   │   ├── mobilenet.py
│   │   ├── resnet.py
│   │   ├── resnet18.py
│   │   ├── resnet18_cbam.py
│   │   ├── resnet18_cbam_concatfusion.py
│   │   ├── resnet18_fusion.py
│   │   ├── samobilenet.py
│   │   ├── seresnet.py
│   │   ├── seresnet18.py
│   │   ├── seresnet18_additionfusion.py
│   │   ├── seresnet18_concatfusion.py
│   │   ├── seresnet18_test_1.py
│   │   ├── seresnet18fusion.py
│   ├── utils/
│       ├── __init__.py
│       ├── custom_losses.py
│       ├── custom_models.py
│       ├── data_manager.py
│       ├── dataset_loader.py
│       ├── eval_metrics.py
│       ├── lr_schedulers.py
│       ├── optimizers.py
│       ├── samplers.py
│       ├── transforms.py
├── .gitignore
├── LICENSE
├── README.md
├── TODO.md
├── args.py
├── main.py
├── mixup_alternate_rows.py
├── output_image.jpg
├── resnet50-19c8e357.pth
├── test.sh
├── train.sh

---

Requirements

• Python 3.8+
• Required libraries:
  • TensorFlow / PyTorch
  • NumPy
  • Matplotlib
  • Additional dependencies (listed in requirements.txt)

Setup

1. Clone the repository:

   git clone <repository-url>
   cd vehicle_reid

2. Install dependencies:

   pip install -r requirements.txt

---

Directory Descriptions

• src/losses/: Contains implementations of custom loss functions used in training models.

  • center_loss.py: Center loss implementation.
  • cross_entropy_loss.py: Cross-entropy loss.
  • hard_mine_triplet_loss.py: Hard-mining triplet loss.

• src/models/: Defines model architectures for vehicle re-identification.

  • resnet18.py: ResNet-18 base architecture.
  • seresnet18.py: SE-ResNet-18 variant.
  • *_fusion.py: Various fusion-based models.

• src/utils/: Utility scripts for data loading, augmentation, optimizers, and evaluation.

  • dataset_loader.py: Handles data loading and preprocessing.
  • eval_metrics.py: Metrics for model evaluation.
  • custom_models.py: Extensions or custom modifications of standard models.

• logs/: Stores logs and checkpoints from training.

• docs/: Documentation for the project.

---

Usage

Training

To train a model:

bash train.sh

Evaluation

To test the model:

python3 main.py --data_path <data-path> --resume <checkpoint-path>

Visualizing Logs

To visualize training logs using TensorBoard:

python3 -m tensorboard.main --logdir logs/

---

Notes

• All checkpoints are stored in the logs/ directory.
• Fusion models (*_fusion.py) implement advanced techniques for feature aggregation.

---

Report

Report can be found at https://github.com/deepwilson/vehicle_reid/blob/main/EEEM071_Coursework_Report.pdf

Feedback on research (Course module leader)

• The research work received the following feedback from the professor: