Skip to content

README.md

Latest commit

 

History

History
315 lines (257 loc) · 9.26 KB

README.md

File metadata and controls

315 lines (257 loc) · 9.26 KB

AI Based Fashion Framework

A comprehensive AI-powered fashion system that combines virtual try-on, fashion recommendation, neural style transfer, color analysis, and preprocessing capabilities. This system provides end-to-end fashion solutions from image preprocessing to personalized recommendations.

Key Features

Virtual Try-On (SCW-VTON)

  • Shape-Guided Clothing Warping: State-of-the-art virtual try-on using the SCW-VTON model
  • Paired/Unpaired Modes: Try on original clothing or different items
  • High-Quality Results: Realistic clothing warping and person synthesis
  • VITON-HD Compatible: Works with VITON-HD datasets

Fashion Recommendation System

  • Text-to-Fashion Search: Find clothing items using natural language descriptions
  • Image-to-Fashion Search: Upload clothing images to find similar items
  • FashionCLIP Integration: Uses state-of-the-art fashion-specific CLIP model
  • FAISS-Powered: Lightning-fast similarity search with vector indexing
  • Multi-modal Queries: Combine text and image searches

Neural Style Transfer

  • Fashion Style Transfer: Apply artistic styles to clothing images
  • Customizable Parameters: Adjustable iterations, weights, and quality settings
  • Background Removal: Optional transparent backgrounds using rembg
  • VGG19-Based: Professional-quality artistic transformations
  • Real-time Processing: Optimized for interactive use

Personal Color Analysis

  • Seasonal Color Matching: Determines whether you're Spring, Summer, Autumn, or Winter
  • Skin Tone Analysis: Uses ResNet-based deep learning for accurate classification
  • Lip Color Extraction: Analyzes facial features for personalized recommendations
  • Color Palette Generation: Provides matching clothing color suggestions
  • Fashion Integration: Links color analysis to clothing recommendations

Comprehensive Preprocessing Pipeline

  • Multi-Stage Processing: Complete fashion image preprocessing workflow
  • DensePose Integration: Human pose and shape estimation
  • OpenPose Support: Skeletal pose detection and JSON generation
  • Image Segmentation: Advanced parsing using PGN (Parsing Graph Network)
  • Mask Generation: Agnostic masks, cloth masks, and binary segmentation
  • Background Removal: Professional cloth mask creation

Web Interface

  • Streamlit Dashboard: User-friendly web interface for all features
  • Multi-Environment Architecture: Isolated conda environments for optimal performance
  • Real-time Processing: Interactive experience with progress tracking
  • Responsive Design: Works on desktop and mobile devices
  • Professional UI: Clean, intuitive interface with custom styling

System Architecture

Fashion AI System
├── Virtual Try-On (SCW-VTON)
│   ├── Step 1: Clothing Warping
│   ├── Step 2: Person Synthesis  
│   └── Evaluation Metrics
├── Fashion Recommender
│   ├── FashionCLIP Embeddings
│   ├── FAISS Vector Database
│   └── Multi-modal Search
├── Neural Style Transfer
│   ├── VGG19 Feature Extraction
│   ├── Style/Content Loss
│   └── Artistic Rendering
├── Color Analysis
│   ├── Face Detection & Parsing
│   ├── Seasonal Classification
│   └── Color Palette Matching
├── Preprocessing Pipeline
│   ├── DensePose Generation
│   ├── OpenPose Extraction
│   ├── Image Segmentation
│   └── Mask Creation
└── Web Interface (Streamlit)
    ├── Multi-tab Interface
    ├── File Upload/Download
    └── Real-time Processing

Quick Start

Prerequisites

  • Python 3.8+
  • CUDA-compatible GPU (recommended for virtual try-on)
  • Miniconda/Anaconda for environment management
  • Git for cloning repositories

1. Clone the Repository

git clone https://github.com/mayank-jangid-moon/Fashion_Shit.git
cd Fashion_Shit

2. Setup Environments

Core Streamlit Environment

cd Streamlit
pip install -r requirements.txt

FashionCLIP Environment (for Recommendations)

cd Streamlit
./setup_fashionclip_env.sh

Color Analysis Environment

cd Streamlit
./setup_color_analysis_env.sh

SCW-VTON Environment (for Virtual Try-On)

cd SCW-VTON
conda create -n scw-vton python=3.8
conda activate scw-vton
bash environment.sh

DensePose Environment (for Preprocessing)

cd Preprocessing/densepose
conda create -n densepose python=3.8
conda activate densepose
pip install -r requirements.txt

3. Download Pre-trained Models

SCW-VTON Models

Download checkpoints from Baidu Netdisk or Google Drive

  • Place in SCW-VTON/ckpts/
  • Download VGG19 checkpoint and place in SCW-VTON/models/vgg/

Color Analysis Models

  • ResNet model should be at: Colour_Analysis/facer/best_model_resnet_ALL.pth

4. Build Fashion Database (Optional)

cd Recommender/FashionCLIP
python build_index.py --data_dir /path/to/your/fashion/images

5. Launch the Application

cd Streamlit
streamlit run app.py

Visit http://localhost:8501 in your browser!

Detailed Usage

Virtual Try-On

cd SCW-VTON
# Step 1: Clothing Warping
python test_for_step1.py --dataroot ./data --ckpt_dir ./ckpts --pair_mode unpaired

# Step 2: Person Synthesis  
python test_for_step2.py --dataroot ./data --ckpt_dir ./ckpts --pair_mode unpaired --plms

# Or use the combined script
bash test.sh

Fashion Recommendations

cd Recommender/FashionCLIP
# Text search
python recommend.py --query "red summer dress" --top_k 10

# Image search  
python recommend.py --image_path /path/to/image.jpg --top_k 10

Neural Style Transfer

cd NST_Clothes
python NeuralStyleTransfer.py \
    --content /path/to/content.jpg \
    --style /path/to/style.jpg \
    --output /path/to/output.jpg \
    --iterations 100

Preprocessing Pipeline

# Full preprocessing pipeline
python preprocessing_pipeline.py \
    --input_image person.jpg \
    --cloth_image shirt.jpg \
    --output_dir ./output/

# Individual preprocessing steps
cd Preprocessing
python agnostic_mask.py --input person.jpg --output mask.png
python cloth_mask.py --input cloth.jpg --output cloth_mask.png

Color Analysis

cd Colour_Analysis
python main.py --input face_photo.jpg --output color_analysis.json

Configuration

Environment Variables

# Set CUDA device (optional)
export CUDA_VISIBLE_DEVICES=0

# Set model paths (if different from defaults)
export SCWVTON_MODEL_PATH=/path/to/scw-vton/models
export FASHIONCLIP_INDEX_PATH=/path/to/faiss/index

Web Interface Settings

Edit Streamlit/.streamlit/config.toml:

[theme]
primaryColor = "#FF6B6B"
backgroundColor = "#FFFFFF"
secondaryBackgroundColor = "#F0F2F6"
textColor = "#262730"

[server]
maxUploadSize = 200

Testing

Run Individual Tests

# Test recommender system
cd Streamlit
python test_recommender.py

# Test style transfer
python test_style_transfer.py

# Test color analysis
python test_color_analysis.py

# Test preprocessing pipeline
cd ..
python preprocessing_pipeline.py --input_image sample_person.jpg --output_dir test_output

Evaluation Metrics

cd SCW-VTON
python metrics.py  # Evaluates FID, SSIM, LPIPS scores

Development

Project Structure

Fashion_Shit/
├── SCW-VTON/                 # Virtual try-on system
│   ├── test_for_step1.py    # Clothing warping
│   ├── test_for_step2.py    # Person synthesis
│   ├── ckpts/               # Model checkpoints
│   └── configs/             # Configuration files
├── Streamlit/               # Web interface
│   ├── app.py              # Main application
│   ├── *_service.py        # Individual services
│   └── setup_*.sh          # Environment setup scripts
├── Recommender/             # Fashion recommendation
│   └── FashionCLIP/        # CLIP-based search
├── NST_Clothes/             # Neural style transfer
├── Colour_Analysis/         # Personal color analysis
├── Preprocessing/           # Image preprocessing
│   ├── densepose/          # Human pose estimation
│   ├── openpose/           # Pose detection
│   └── image_parse/        # Image segmentation
└── Dataset/                # Training/test data

Research & Citations

This project combines several state-of-the-art research works:

SCW-VTON (Virtual Try-On)

@inproceedings{han2024shape,
  title={Shape-Guided Clothing Warping for Virtual Try-On},
  author={Han, Xiaoyu and Zheng, Shunyuan and Li, Zonglin and Wang, Chenyang and Sun, Xin and Meng, Quanling},
  booktitle={Proceedings of the 32nd ACM International Conference on Multimedia},
  pages={2593--2602},
  year={2024}
}

FashionCLIP (Recommendations)

  • Based on OpenAI CLIP adapted for fashion domain
  • Combines text and image understanding for fashion

Neural Style Transfer

  • Based on Gatys et al. "A Neural Algorithm of Artistic Style"
  • Uses VGG19 for feature extraction and style transfer

Team

  1. Mayank Jangid
  2. Abhinav Rajput
  3. Odwitiyo Dutta