AI-Powered Multimodal Product Price Estimation
Estimate product prices using product descriptions, images, and structured product attributes through a multimodal machine learning pipeline.
PriceVision AI
https://multimodal-price-predictor.streamlit.app/
PriceVision AI is a multimodal machine learning application that estimates product prices by combining Natural Language Processing (NLP), Computer Vision, and structured product attributes.
The application analyzes:
- Product descriptions
- Product images
- Structured product attributes
to generate:
- Estimated Market Price (USD)
- 🇮🇳 Approximate INR Conversion
- Downloadable PDF Report
- Multimodal AI price estimation
- TF-IDF text feature extraction (50,000 features)
- EfficientNetB0 image embeddings (1,280 features)
- Structured feature engineering
- 51,282-dimensional feature fusion
- Optimized XGBoost regression model
- Estimated Market Price (USD)
- 🇮🇳 Approximate INR conversion
- PDF report generation
- Modern responsive Streamlit interface
- Cached inference for faster predictions
Product Description
│
▼
TF-IDF (50,000 Features)
Product Image ─────► EfficientNetB0 (1,280 Features)
Structured Attributes
(weight, quantity, pack size)
│
▼
Feature Fusion (51,282 Features)
│
▼
Optimized XGBoost Model
│
▼
Estimated Market Price (USD)
│
▼
Approximate INR Conversion
PriceVision-AI
│
├── app/ Streamlit application
├── backend/ Model loading and inference
├── components/ Reusable UI components
├── assets/ Images and static assets
├── data/ Runtime model files
├── notebooks/ Model development notebooks
├── screenshots/ README images
├── src/ Experimental utilities
├── styles/ Global styling
├── requirements.txt
└── README.md
| Category | Technology |
|---|---|
| Language | Python |
| Frontend | Streamlit |
| NLP | TF-IDF |
| Computer Vision | TensorFlow, EfficientNetB0 |
| Machine Learning | XGBoost |
| Image Processing | Pillow |
| Data Processing | Pandas, NumPy |
| Utilities | Scikit-learn, Joblib |
Clone the repository.
git clone https://github.com/Urvity03/Multimodal-Product-Price-Predictor.gitNavigate to the project.
cd Multimodal-Product-Price-PredictorInstall the dependencies.
pip install -r requirements.txtRun the application.
streamlit run app/app.pyNote: The first prediction may take longer while the TF-IDF vocabulary and EfficientNetB0 resources initialize. Subsequent predictions reuse cached resources for faster inference.
| Component | Purpose |
|---|---|
| TF-IDF | Product description features |
| EfficientNetB0 | Image feature extraction |
| Structured Feature Engineering | Weight and pack-size extraction |
| XGBoost | Final multimodal regression model |
The deployed application reconstructs the TF-IDF vocabulary deterministically from the original ordered corpus to preserve compatibility with the trained XGBoost model. The prediction model itself is not retrained during inference.
This model was trained on a broad e-commerce dataset comprising approximately 5,000 curated products.
Predictions are intended to provide intelligent market estimates based on product descriptions, images, and structured attributes. While the model performs well for products similar to those represented in the training data, estimates for premium, newly released, or niche products may vary due to limited training coverage.
| Metric | Value |
|---|---|
| Model | Optimized Multimodal XGBoost |
| Mean Absolute Error (MAE) | 13.86 |
| Output | Estimated Market Price (USD) |
Note: Predictions are intended as intelligent market estimates rather than exact retail prices.
- Category-specific pricing models
- Brand-aware feature engineering
- Larger multimodal training datasets
- Confidence estimation for predictions
- Multi-currency support
- Live exchange-rate integration
- Model explainability (SHAP/LIME)
- Performance optimization and MLOps integration
Urvi Tyagi
B.Tech – Artificial Intelligence & Machine Learning
- GitHub: https://github.com/Urvity03
- LinkedIn: https://www.linkedin.com/in/urvi-tyagi-17b302286/
Project Repository
https://github.com/Urvity03/Multimodal-Product-Price-Predictor
This project is licensed under the MIT License.
⭐ If you found this project useful, consider giving it a star on GitHub!