towhee.functional.mixins.faiss.FaissMixin

class towhee.functional.mixins.faiss.FaissMixin[source]

Bases: object

Mixins for Faiss, such as loading data into Faiss. And ids and vectors need to be passed as index. If ids is a string, KV storage will be started, and the kv data will be saved to the specified directory as “kv.bin”.

Parameters:

  • findex (str or faiss.INDEX, optional) – The path to faiss index file(defaults to ‘./index.bin’) or faiss index.

  • string (str, optional) – A string to produce a composite Faiss index, which is the same parameter in faiss.index_factor, defaults to ‘IDMap,Flat’, and you can refer to https://github.com/facebookresearch/faiss/wiki/The-index-factory.

  • metric (faiss.METRIC, optional) – The metric for Faiss index, defaults to faiss.METRIC_L2.

Returns:

A DC, and will save the Faiss index file and kv file(if ids is string).

Examples:

Note

Please make sure the path to index_file is authorized, and it will write the Faiss index file and kv file(if ids is string).

>>> import towhee
>>> dc = ( 
...     towhee.glob['path']('./*.jpg')
...           .image_decode['path', 'img']()
...           .image_embedding.timm['img', 'vec'](model_name='resnet50')
...           .to_faiss['path', 'vec'](findex='./faiss/faiss.index')
... )

Methods

__init__()[source]