semantic text2image search

Semantic Text2Image Search

This code let's you search your images in a descriptive manner.

Usage

1. Installation

In a location where you want to place this repo, use the following commands:

git clone [email protected]:bauerem/semantic-text2image-search.git
cd semantic-text2image-search
source setup.sh

2. Add Your Own Data

i. Place your images into the backend/images directory.

ii. Make them searchable with the following script:

source vectorize.sh

3. Search

You can now run the software using the command:

source run_backend.sh

And in another terminal, using the command:

source run_frontend.sh

Modifications depending on use-case

• Set the standard embedding function to CLIPProcessor so that Chroma embeds directly
• Tokenization of the query can be sped up by using the CLIPTokenizerFast class in embed.py
• Adjust the 'where_document' parameter in the collection query to only search relvant