Image Similarity Search

Search for similar images using Vertex AI's multimodal embeddings and Faiss.

How it works

1. It starts by getting all the images from the bucket specified in the config.storage_bucket variable.
2. It then loops through all of the files and generate embeddings for them using Vertex AI's multimodal embeddings.
3. The embeddings are then saved in a file which can be modified through the config.filepath variable.
4. The UI is made using Streamlit, when you upload an image in the Streamlit app the script converts the image into embeddings.
5. Then a Faiss index is created to perform similarity search over the previously saved embeddings and the embeddings created from the image
6. The result is then displayed in the Streamlit app

Note: The config.index_dimensions variable is set by default to 1408 because the Embeddings for Multimodal multimodalembedding model generates 1408 dimensions vectors.

Installation

1. Clone the repository:

git clone https://github.com/ItzCrazyKns/Image-Search.git

2. Modify the config.json file.
3. Install required packages:

pip install -r requirements.txt

4. Generate embeddings for the images saved in the bucket:

python generateEmbeddings.py

5. Run the Streamlit app:

streamlit run main.py

Note: Make sure you're authenticated in the gcloud CLI.