Skip to content

README.md

Latest commit

 

History

History
245 lines (167 loc) · 5.96 KB

README.md

File metadata and controls

245 lines (167 loc) · 5.96 KB
Raw

PriceVision AI

AI-Powered Multimodal Product Price Estimation

Estimate product prices using product descriptions, images, and structured product attributes through a multimodal machine learning pipeline.

Python Streamlit TensorFlow XGBoost Status

Live Demo

PriceVision AI

https://multimodal-price-predictor.streamlit.app/

Overview

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

Key Highlights

  • 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

Application Preview

Home Page

Home

Prediction Workspace

Prediction Workspace

Results Dashboard

Results Dashboard

Model Architecture

                 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

Project Structure

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

Technology Stack

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

Installation

Clone the repository.

git clone https://github.com/Urvity03/Multimodal-Product-Price-Predictor.git

Navigate to the project.

cd Multimodal-Product-Price-Predictor

Install the dependencies.

pip install -r requirements.txt

Run the application.

streamlit run app/app.py

Note: The first prediction may take longer while the TF-IDF vocabulary and EfficientNetB0 resources initialize. Subsequent predictions reuse cached resources for faster inference.

Model

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.

Prediction Scope

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.

Performance

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.

Future Improvements

  • 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

Developer

Urvi Tyagi

B.Tech – Artificial Intelligence & Machine Learning

Project Repository

https://github.com/Urvity03/Multimodal-Product-Price-Predictor

License

This project is licensed under the MIT License.

⭐ If you found this project useful, consider giving it a star on GitHub!