# Color tags

## An App classifying images on different colors 

App: https://huggingface.co/spaces/rrighart/color-tags

Color tags are very useful for finding pictures on the web, such as in e-commerce applications.
This App provides color tags for uploaded image files.
The underlying model was trained using a basic Convolutional Neural Network and an EfficientNet, with the latter giving slightly better results [[1]](https://www.kaggle.com/code/rrighart/the-prediction-of-color-tags/data) .

For training and validation, the Vehicles Color Recognition (VCoR) dataset was used, which has 15 colors classes [[2]](https://www.kaggle.com/datasets/landrykezebou/vcor-vehicle-color-recognition-dataset). A separate unseen testset has overall accuracy of 84%.
One goal of this project is to see if the model can generalizes to other kind of products and objects. For this purpose tested some images from the Fashion Product Images data [[3]](https://www.kaggle.com/datasets/paramaggarwal/fashion-product-images-small) which shows promising results.

## Contact

Ruthger Righart (PhD)

Data scientist in machine learning and computer vision

Email: rrighart@googlemail.com

Web: https://www.rrighart.com

## References

1. The prediction of color tags, Kaggle Notebook R. Righart.
2. Vehicle color recognition dataset (VCoR).
3. Fashion Product Images.