This repository contains the research paper supplementary materials for our study on body and garment parts detection and analysis utilizing computer vision techniques. Our work introduces a novel approach to identify and segment specific body parts and clothing items within images, which is vital for understanding current fashion trends and assisting in the creation of fashion recommendation systems.
Proposed model architecture
Pixel intensity evaluation for person localisation.
The core of our methodology is a multi-stage process that includes:
The study utilizes several publicly available datasets, such as Fashionpedia, DARN, and Fashion10000, providing a diverse range of images for analysis. Real-time videos and images from sources like Pexels are also included to demonstrate the model's practical applications.
To run this project, you will need to change the following variables to your config.py file
harcascade_classifier_path
# Path where haarcascade is resides.
cnn_model_path
# Path where Fully Connected pre-trained model weights resides.
yolo_face_detection_model_path
# Path where yolo-v9 pre-trained face-detection model weights resides.
yolo_person_detection_model_path
# Path where yolo-v9 person-detection pre-trained model weights resides.
device=torch.device("cpu")
# change to gpu if you have gpu enabled devide.
TEMP_DIR
# Directory to store temporary files.
EVAL_DIR
# Directory to store evaluation files.
RAW_DIR_NAME
# Raw files while using YOLOv9
Python 3.9
pip install -r requirement.txt
python main.py
Comparison results generated by our model with Haarcascade and YOLOv9 with Fully-connected CNN.
Real time results.
Our findings indicate a significant improvement in the detection of fashion-related elements in images, with a particular enhancement noted in the accuracy of face detection and the subsequent body part segmentation.
This research contributes to the advancement of fashion-oriented computer vision applications, demonstrating the potential for automated analysis of fashion elements in images. The methods developed herein hold promise for diverse applications, from trend prediction to retail strategy and security.
@article{wang2024yolov9,
title={{YOLOv9}: Learning What You Want to Learn Using Programmable Gradient Information},
author={Wang, Chien-Yao and Liao, Hong-Yuan Mark},
booktitle={arXiv preprint arXiv:2402.13616},
year={2024}
}