# Person-Re-Identification

**Repository Path**: greitzmann/Person-Re-Identification

## Basic Information

- **Project Name**: Person-Re-Identification
- **Description**: Person re-identification, a tool used in intelligent video surveillance, is the task of correctly identifying individuals across multiple images captured under varied scenarios from multiple cameras. Solving this problem is inherently a challenging one because of the issues posed to it by low resolution images, illumination changes per image, unconstrained pose and occlusions.
- **Primary Language**: Unknown
- **License**: MIT
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2021-01-27
- **Last Updated**: 2021-01-27

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# Person-Re-Identification
For making sense of the vast quantity of visual data generated by the rapid expansion of large-scale distributed multi-camera systems, automated person re-identification is essential. Person re-identification, a tool used in intelligent video surveillance, is the task of correctly identifying individuals across multiple images captured under varied scenarios from multiple cameras. Solving this problem is inherently a challenging one because of the issues posed to it by low-resolution images, illumination changes per image, unconstrained pose and occlusions. 

In this project, we aim at developing a Person re-identification model using Deep Neural Networks (DNN) which can handle variable size input images. Specifically, we aim at implementing two preprocessing techniques, which reduce the chances of overfitting i.e. we aim to make our model robust to occlusion using Random Erasing, a data augmentation technique, and reduce the influence of pose variations on features using a Pose normalized Generative Adversarial Network (GAN). 

Along with this we also aim to implement and integrate Part-based Convolutional Baseline (PCB) to further improve on the results. We briefly describe the models trained along with their evaluation results on Market1501 dataset and provided validation and test sets.

# Baseline used 
https://github.com/KaiyangZhou/deep-person-reid

# Pose-Normalization
https://github.com/naiq/PN_GAN.git

Paper Link : http://openaccess.thecvf.com/content_ECCV_2018/papers/Xuelin_Qian_Pose-Normalized_Image_Generation_ECCV_2018_paper.pdf

<b>READ THIS FOR MORE INFORMATION : </b> [Pose Normalized Training](https://github.tamu.edu/aditya30394/Person-Re-Identification/blob/master/deep-person-reid/README.md)

# Model-Outline
<p align="center">
<figure align="center">
  <img src="https://github.com/aditya30394/Person-Re-Identification/blob/master/TrainingBlockSchematic.PNG">
  <figcaption>Training Block Schematic</figcaption>
</figure>  
  <br/>
</p>


# GAN result
Output after 10th epoch <br>
![Result after 10 epochs](https://github.com/aditya30394/Person-Re-Identification/blob/master/GAN_training_epoch_10.png)

Output after 12th epoch <br>
![Result after 12 epochs](https://github.com/aditya30394/Person-Re-Identification/blob/master/GAN_training_epoch_12.png)

# Random Erasing Data Augmentation
https://github.com/zhunzhong07/Random-Erasing

Paper Link : https://arxiv.org/abs/1708.04896

# Link to the trained model(only erasing)
https://drive.google.com/open?id=1Gm7hpF3HoG2Xt0WV92Wi07U428BH8B1q

# Links to the feature mat files(according to market1501)
Query: https://drive.google.com/open?id=1StnqZt9MOqiyUYnf_RfhBGXHWQiFgLpz

Gallery: https://drive.google.com/open?id=1jeoQyxqtRW07M1Shbe4pt9Aw3e-m1icY
# Result feature .mat files
The extracted features on the test Set are in the folder Result_mat, where the three result set feature files (feature_test_query.mat and feature_test_gallery.mat) are in Result_1, Result_2, Result_3 respectively.

## License

This project is licensed under the MIT License - see the [LICENSE](LICENSE) file for details