Food Taste Similarity Prediction Based on Images and Human Judgments
![](/project/eth-iml-4/featured_hu964e47df092987bdf0a35da0cb778e6b_1611832_02da8237b8555e8dab3f64120ae77128.webp)
Description
A dataset of images of 10,000 dishes, together with a set of triplets representing human annotations of food taste similarity is provided. The objective of the project is to predict for unseen triplets $(A, B, C)$, whether dish $A$ is more similar in taste to $B$ or $C$. In this project, image embeddings are learned through an autoencoder, where the loss is computed based on distances between the embeddings.