The goal of this repository is to search for images on Steam using natural language queries.
The dataset consists of 29,982 vertical images (300x450 resolution), downloaded from Steam.
Image features have been pre-computed, using OpenAI's CLIP model based on Vision Transformer architecture.
Below are shown results for:
- image search, i.e. using natural language queries,
- reverse image search, i.e. using image queries.
steam-CLIP: retrieve games using image queries,- interactive exploration with a web app using an appID,
- interactive exploration with a web app using a dropdown menu.
steam-image-search: retrieve games using natural language queries,