This project aims to classify products into their correct categories using computer vision techniques. Accurate categorization is crucial for e-commerce companies to enhance customer satisfaction and optimize sales. The model developed in this project helps reduce misclassification, increase sales, and improve the efficiency of the product categorization process.
This study is not a commercial product, rather an experimental progress and intended to contribute to open-source data science community.
We have collected ecommerce product images from various resources such as Amazon, Walmart, Google and other sources through scraping. Also Amazon Berkeley Objects (ABO) project used to enhance images in some categories. Finally we have reached a dataset of 18K images, properly resized into 224*224 pixels which is suitable for many pretrained CNN models. We used PILLOW for image processing and selenium for web scraping.
18,175 images categorized in 9 classes mostly due to their major categories in Amazon. Furthermore, train-val-check sets arranged appropriately and train-val sets used for model training and a small check set reserved for model deployment to evaluate visually the model performance. The images is available on static folder
The notebook for the study can be found at kaggle
-
ImageDataGenerator: Applied data augmentation techniques such as rotation, horizontal/vertical flipping, zoom, and brightness adjustments using Keras' ImageDataGenerator class to increase variation in the dataset and reduce overfitting.
-
Pre-trained Models: Used transfer learning with models like VGG19, VGG16, ResNet, MobileNetV2, EfficientNet, and InceptionV3. The best performing model be seen as MobileNetV2
-
Optimization Algorithms: Tested different optimization algorithms such as Adam and RMSprop to improve accuracy and reduce loss.
Flask and docker was used to deploy the model, enabling real-time interaction and visualization of the model's results. This streamlined the process of integrating the model into real-world applications.
Ensure you have the following installed:
- Python 3.7+
- pip (Python package installer)
git clone https://github.com/fatih-ml/ecommerce_images_CV_classification/.git
cd ecommerce_images_CV_classification
python -m venv venv
source venv/bin/activate # On Windows use `venv\Scripts\activate`
pip install -r requirements.txt
flask run
Open your web browser and go to http://127.0.0.1:5000/ to interact with the app.
We welcome contributions and suggestions! Feel free to fork this repository and submit pull requests.
For any questions or further information, you can reach out to me on LinkedIn.
Thanks for reading and visiting our repo! We hope this project serves as a helpful educational resource.