ripple-net 0.1.3

Text-image search and image tagging library

ripple_net (wip)

A library for text/image based search/retrieval for image datasets and files. Uses multimodal AI techniques/models like vector embeddings and CLIP.

Install

$ pip install ripple_net

Usage

• For text description-based search

from ripple import ImageEmbedder, TextSearch # import classes

# load from a huggingface image dataset or load from a local image directory
embedder = ImageEmbedder('huggan/wikiart', retrieval_type='text-image', dataset_type='huggingface') 

# could also use 'cpu' if CUDA-enabled GPU isn't available
embedded_images = embedder.create_embeddings(device="cuda", batch_size=32)

# initialize text - image search class
text_search = TextSearch(embedded_images, embedder.embed_model)

# specify text/search query for image, and number of results to return
scores, images = text_search.get_similar_images(query='painting of a river', k_images=10) 

images['image'][0].show()

• For image-based retrieval(image-image search)

from ripple import ImageEmbedder, ImageSearch, image_loader

 # load dataset and initialize embedding class
embedder = ImageEmbedder('lambdalabs/naruto-blip-captions', retrieval_type='image-image', dataset_type='huggingface')

# generate embeddings
embedded_images = embedder.create_embeddings(device="cuda", batch_size=32)

# init image search class
image_search = ImageSearch(embedded_images, embedder.embed_model)

# retrieve similar images with image input
input_image = image_loader('katara.png') # use library function to load image in PIL format

scores, images = image_search.image_search(input_img=input_image, k_images=5) # specify input image, and number of results to return

# dislay one of retrieved images
images['image'][0].show()
# or using notebooks => images['image'][0]

Todo

coming soon

• add auto-image file tagging/renaming
• direct CLI usage

Acknowledgement

• Image search engine: article by not-lain
• CLIP (Contrastive Language–Image Pre-training) research by OpenAI.