README.md•7.14 kB
# FastMCP_RecSys
This is a CLIP-Based Fashion Recommender with MCP.
### 📌 Sample Components for UI
1. Image upload
2. Submit button
3. Display clothing tags + recommendations
# Mockup
A user uploads a clothing image → YOLO detects clothing → CLIP encodes → Recommend similar
<img width="463" alt="Screenshot 2025-04-26 at 10 26 13 AM" src="https://github.com/user-attachments/assets/93c0a75b-4ed1-4fa1-b25d-5137b8eb6af0" />
# Folder Structure
```
/project-root
│
├── /backend
│ ├── Dockerfile
│ ├── /app
│ ├── /aws
│ │ │ └── rekognition_wrapper.py # AWS Rekognition logic
│ │ ├── /utils
│ │ │ └── image_utils.py # Bounding box crop utils
│ │ ├── /controllers
│ │ │ └── clothing_detector.py # Coordinates Rekognition + cropping
│ │ ├── /tests
│ │ │ ├── test_rekognition_wrapper.py
│ │ │ └── test_clothing_tagging.py
│ │ ├── server.py # FastAPI app code
│ │ ├── /routes
│ │ │ └── clothing_routes.py
│ │ ├── /controllers
│ │ │ ├── clothing_controller.py
│ │ │ ├── clothing_tagging.py
│ │ │ └── tag_extractor.py # Pending: define core CLIP functionality
│ │ ├── schemas/
│ │ │ └── clothing_schemas.py
│ │ ├── config/
│ │ │ ├── tag_list_en.py $ Tool for mapping: https://jsoncrack.com/editor
│ │ │ ├── database.py
│ │ │ ├── settings.py
│ │ │ └── api_keys.py
│ │ └── requirements.txt
│ └── .env
│
├── /frontend
│ ├── Dockerfile
│ ├── package.json
│ ├── package-lock.json
│ ├── /public
│ │ └── index.html
│ ├── /src
│ │ ├── /components
│ │ │ ├── ImageUpload.jsx
│ │ │ ├── DetectedTags.jsx
│ │ │ └── Recommendations.jsx
│ │ ├── /utils
│ │ │ └── api.js
│ │ ├── App.js # Main React component
│ │ ├── index.js
│ │ ├── index.css
│ │ ├── tailwind.config.js
│ │ └── postcss.config.js
│ └── .env
├── docker-compose.yml
└── README.md
```
## Quick Start Guide
### Step 1: Clone the GitHub Project
### Step 2: Set Up the Python Environment
```
python -m venv venv
source venv/bin/activate # On macOS or Linux
venv\Scripts\activate # On Windows
```
### Step 3: Install Dependencies
```
pip install -r requirements.txt
```
### Step 4: Start the FastAPI Server (Backend)
```
uvicorn backend.app.server:app --reload
```
Once the server is running and the database is connected, you should see the following message in the console:
```
Database connected
INFO: Application startup complete.
```
<img width="750" alt="Screenshot 2025-04-25 at 1 15 45 AM" src="https://github.com/user-attachments/assets/7f3fc403-fb33-4107-a00c-61796a48ecec" />
### Step 5: Install Dependencies
Database connected
INFO: Application startup complete.
```
npm install
```
### Step 6: Start the Development Server (Frontend)
```
npm start
```
Once running, the server logs a confirmation and opens the app in your browser: [http://localhost:3000/](http://localhost:3000/)
<img width="372" alt="Screenshot 2025-04-25 at 9 08 50 PM" src="https://github.com/user-attachments/assets/794a6dba-9fbb-40f1-9e57-c5c2e2af1013" />
# What’s completed so far:
1. FastAPI server is up and running (24 Apr)
2. Database connection is set up (24 Apr)
3. Backend architecture is functional (24 Apr)
4. Basic front-end UI for uploading picture (25 Apr)
## 5. Mock Testing for AWS Rekognition -> bounding box (15 May)
```
PYTHONPATH=. pytest backend/app/tests/test_rekognition_wrapper.py
```
<img width="1067" alt="Screenshot 2025-05-20 at 4 58 14 PM" src="https://github.com/user-attachments/assets/7a25a92d-2aca-42a8-abdd-194dd9d2e8a5" />
- Tested Rekognition integration logic independently using a mock → verified it correctly extracts bounding boxes only when labels match the garment set
- Confirmed the folder structure and PYTHONPATH=. works smoothly with pytest from root
## 6. Mock Testing for AWS Rekognition -> CLIP (20 May)
```
PYTHONPATH=. pytest backend/app/tests/test_clothing_tagging.py
```
<img width="1062" alt="Screenshot 2025-05-21 at 9 25 33 AM" src="https://github.com/user-attachments/assets/6c64b658-3414-4115-9e20-520132605cab" />
- Detecting garments using AWS Rekognition
- Cropping the image around detected bounding boxes
- Tagging the cropped image using CLIP
## 7. Mock Testing for full image tagging pipeline (Image bytes → AWS Rekognition (detect garments) → Crop images → CLIP (predict tags) + Error Handling (25 May)
| **Negative Test Case** | **Description** |
| -------------------------------| ------------------------------------------------------------------------------- |
| No Detection Result | AWS doesn't detect any garments — should return an empty list. |
| Image Not Clothing | CLIP returns vague or empty tags — verify fallback behavior. |
| AWS Returns Exception | Simulate `rekognition.detect_labels` throwing an error — check `try-except`. |
| Corrupted Image File | Simulate a broken (non-JPEG) image — verify it raises an error or gives a hint. |
```
PYTHONPATH=. pytest backend/app/tests/test_clothing_tagging.py
```
<img width="1072" alt="Screenshot 2025-05-21 at 11 19 47 AM" src="https://github.com/user-attachments/assets/b41f07f4-7926-44a3-8b64-34fe3c6ef049" />
- detect_garments: simulates AWS Rekognition returning one bounding box: {"Left": 0.1, "Top": 0.1, "Width": 0.5, "Height": 0.5}
- crop_by_bounding_box: simulates the cropping step returning a dummy "cropped_image" object
- get_tags_from_clip: simulates CLIP returning a list of tags: ["T-shirt", "Cotton", "Casual"]
## 8. Run Testing for CLIP Output (30 May)
```
python3 -m venv venv
pip install -r requirements.txt
pip install git+https://github.com/openai/CLIP.git
python -m backend.app.tests.test_tag_extractor
```
<img width="1111" alt="Screenshot 2025-06-06 at 5 12 13 PM" src="https://github.com/user-attachments/assets/d0b3b288-20f8-482f-9d39-dcccf9a775ee" />
Next Step:
1. Evaluate CLIP’s tagging accuracy on sample clothing images
2. Fine-tune the tagging system for better recommendations
3. Test the backend integration with real-time user data
4. Set up monitoring for model performance
5. Front-end demo