/vehant-scs-par

Primary LanguagePython

Person Attribute Recognition

This project aims to perform Person Attribute Recognition (PAR) using a pretrained model and fine tunned on downstream task. Below are the steps to set up and run the project.
The main branch contains the pipeline for FEAT (Feature Classifier with ResNet50 Backbone) model (best performing model).
The swin branch contains the pipeline for SWIN (Shifted Window Transformer) model (second best performing model).
The beit branch contains the pipeline for BEiT (Bidirectional Encoder representation from Image Transformers) model (least performing model).

Requirements

Before running the code, install the necessary dependencies by running:

pip install -r requirements.txt

Steps to Run the Project

1. Data Augmentation

Run the data augmentation script to preprocess and augment the training images:

train and val folder contain the image which will be augmented for further processing

python augmentation.py

2. Utility Functions

Execute the labels functions script which contains code to map label to augmented images and save in csv functions for the project:

python labels.py

3. Model Training

Train the model : Important to Note is we train GPU so cuda is enabled

python train.py

4. Inference

Perform inference on the test data using the trained model:

python inference.py

Summary

  1. Augmentation of the training images was performed.
  2. A pretrained model from github was used.
  3. The model was fine-tuned on the downstream task of Person Attribute Recognition (PAR).
  4. The model was then tested on the provided test data.

Files

  • requirements.txt: Lists the required packages.
  • augmentation.py: Script for data augmentation.
  • labels.py: Contains helper functions.
  • train.py: Script for training the model.
  • inference.py: Script for performing inference on the test data.
  • submision.txt: prediction on test data
  • best_model_v2.pth: model trained and saved.
  • best_model.pth: pretrained used for PAR task.

Installation

To install the required packages, run:

pip install -r requirements.txt

Disclaimer: Shared dataset annotation is the legal property of Vehant Technology; strictly for challenge use only – refrain from using it for any other application or publication, ensure data security, and delete after challenge completion.

The dataset is Not shared Because of the above Disclaimer , still you can have look into code for reference.

Contact

For any questions or issues, please contact

Shubham kale shubham23094@iiitd.ac.in

Shashank Sharma shashank23088@iiitd.ac.in

Abhilash Khuntia abhilash23007@iiitd.ac.in

This README provides an overview of the project structure and instructions for running each step. Make sure to follow the steps in order to ensure proper execution of the project.