deepfates/memery

Search over large image datasets with natural language and computer vision

What it solves

Memery solves the problem of manually searching through large folders of images where filenames are unhelpful or unknown. Instead of scrolling through hundreds of thumbnails, users can find specific images using natural language descriptions.

How it works

The project uses OpenAI's CLIP (Contrastive Language-Image Pretraining) transformer, which maps both images and text into the same latent space. This allows the system to compare a text query (e.g., "a line drawing of a woman facing to the left") against the images in a local folder to find the most relevant matches.

Who it’s for

It is designed for anyone with large local image collections—such as memes, screenshots, datasets, or product photos—who needs a way to search them without relying on filenames or dates.

Highlights

  • Multiple Interfaces: Offers a browser-based GUI, a command-line interface (CLI), and a Python library for integration into other workflows.
  • Flexible Querying: Supports searching via text queries, image queries, or a combination of both.
  • Indexing: Includes a build command to encode and index images for more efficient searching.
  • Notebook Compatible: Can be used within Jupyter notebooks, including its GUI functions.