Multi class object classification using Freiburg Groceries Dataset

Check out the corresponding medium blog post https://towardsdatascience.com/multi-class-object-classification-for-retail-products-aa4ecaaaa096.

The Freiburg Groceries Dataset consists of 5000 256x256 RGB images of 25 food classes. I have used Transfer Learning using pre trained VGG weights to make a model for multi class image classification. It can be of great use in automatic image classification to help a food and groceries retailer reduce human effort in the inventory management process of its warehouse and retail outlets.

The paper can be found here and the dataset here. The full dataset is provided in this repository for convenience.

Environment and tools

1. Jupyter Notebook
2. Numpy
3. Pandas
4. Scikit-image
5. Matplotlib
6. Scikit-learn
7. Keras

Installation

pip install numpy pandas scikit-image matplotlib scikit-learn keras

jupyter notebook

Results

Loss/Accuracy vs Epoch

ROC-AUC curve

Correct/Incorrect classification samples

The model is able to reach a validation accuracy of 60% which is quite good considering the number of classes(25) with 100-200 images in each category. Feel free to use different architectures and play with the hyperparameters for better results.

Citing

@misc{Abhinav:2019,
  Author = {Abhinav Sagar},
  Title = {Grocery Image Classification},
  Year = {2019},
  Publisher = {GitHub},
  Journal = {GitHub repository},
  Howpublished = {\url{https://github.com/abhinavsagar/Grocery-Image-Classification}}
}