A Custom Convolutional Neural Network Architecture based Recommendation System for Taneira, Titan.
This solution was one among 35 selected projects out of 250 projects showcased during the company's "Technology Day".
The dataset (images of sarees, SKU details, and attributes file) are proprietary data of Titan Company Limited, and therefore have been omitted from this project repository. They have also been removed from the results images as discussed below.
- To design a deep learning model to recognise patterns in sarees.
- By recognising patterns in sarees the model should be able to deliver similar sarees from database, which can be used in various use cases.
- We would require the model to predict each characteristic feature/attributes (eg: zari, pallu, colour etc) individually and use the results collectively.
- A saree has multiple attributes, Origin, Weft, Warp, Colour, Loom, Zari, Pallu amongst others.
- This POC we will look into two attributes that have been implemented- Zari (Present or Not) and Pallu (Contrast or Running).
- The user is allowed to enter the colour of the saree they desire.
One can see here in 6th prediction above, where the database entry was made wrong, although the prediction was accurate.
