InsightXtract is a Flask-based web application designed to extract entity types (e.g., height, weight) and their corresponding values (e.g., 50g, 180cm) from product images. The app uses a trained machine learning model (using neural network) for entity extraction and value prediction.
- Extracts entity types like
weight,height, etc., and their values from uploaded product images. - Uses a pre-trained model
EfficientNetB0to create our own trained model (model22f.keras) for prediction. - Provides sample test images (
test1.jpgandtest2.jpg) for users to test the functionality. - Built on the Flask framework with a simple, user-friendly interface.
-
The dataset used for training includes:
- CSV Files: Metadata and labels for model training.
- NPZ Files: Containing grouped data used to create the
mergedData.npzfile. - The
mergedData.npzfile is the primary input for the model training process.
-
A Google Drive link is available to download the dataset, which includes both CSV and NPZ files. Ensure you download and place them in the appropriate folder before running the training scripts.
- The trained model (
model22f.keras) is used for predictions in the application. - It was created using the notebook
model2UsingPreTrained.ipynb. - This model is stored in the
savedModel/folder for easy access during runtime.
- A newer, experimental model is being developed using the notebook
BetaModel3UsingPretrained.ipynb. This model aims to enhance prediction accuracy and overall performance. - The improved model is not yet integrated into the application.
- Install Requirements: Ensure you have Python 3.8+ and necessary Python packages (Flask, TensorFlow/Keras, NumPy, etc.) installed.
- Run the Application:
- Navigate to the
websitefolder and run the Flask app. - Access the app in your browser at
http://127.0.0.1:5000.
- Navigate to the
- Test the Application:
- Use the sample test images (
test1.jpgandtest2.jpg) provided in the project. - Or upload your own product images for predictions.
- Use the sample test images (
Here are some screenshots demonstrating the application's output:
- Model Improvements: Integrate the newer model once it’s finalized.
- Authentication: Add a secure login and authentication system for the application.
- Cloud Integration: Migrate storage to cloud platforms like AWS S3 or Google Cloud for scalability.
- Dataset Expansion: Add more diverse data for improved model generalization.
This project is licensed under the MIT License. See the LICENSE file for more details.
For any queries or contributions, feel free to reach out:
- Author: Anmol Chadha
- Email: [email protected]