Image vector embeddings

Use CLIP a multi-modal vectorizer

Our default vectorizer is using CLIP from OpenAI. CLIP vectorize image or text into same space. This enables functionalities such as text to image search.
Lets first install clip first.

pip install git+https://github.com/openai/CLIP.git

Once installed you can vectorize any image fields. You vectorize multiple fields at once. Your image fields needs to be a url of an image file. If your images are in your local file system, upload those files first with Inserting files.

image_fields = ["image_url"]
ds.vectorize_image(fields=image_fields)

Use BIT from TFHub

A image model from Google. Great for reverse image search.

ds.vectorize_image(fields=["thumbnail"], models=["m-r50x1"])

Filter and Vectorize

You can also filter the images you want vectorized.

ds.vectorize_image(fields=image_fields, filters=filters)