Cannot use operator towhee/to_faiss while using towhee 0.9.0 :-(

[LiuersNick]

I am trying to use Faiss to build an index on the Embedding of images:

import towhee

img_vectors = (
    towhee.read_csv('info.csv')
      .image_decode['file_name', 'img']()
      .image_text_embedding.clip['img', 'vec'](model_name='clip_vit_base_patch32', modality='image')
      .tensor_normalize['vec','vec']() # normalize vector
      .select['file_name', 'vec']()
)

img_vectors.to_faiss['file_name', 'vec'](findex='./index.bin')

but an error occured:

ValueError: operator towhee/to-faiss not found!

My environements look like this:

• python 3.9
• ipykernel 6.15
• towhee 0.9.0

p.s. I understand that the operator towhee/to_faiss has been added in version 0.7.1 (it is written in README.md, branch main), but I don’t know why this error is reported

Thanks!

[junjiejiangjjj]

@Chiiizzzy Please take a look

[Chiiizzzy]

to_faiss is not an operator, it's a mixin. Could you please try to run from towhee.functional.mixins.faiss import FaissMixin and post the result here. I tried to reproduce your issue, but it seems things work smoothly in my new conda env.

[LiuersNick]

I tried to reproduce your issue, but it seems things work smoothly in my new conda env.

Thanks for your reply! I tried to create a new conda env and only install towhee, gradio and faiss-cpu and rerun

img_vectors.to_faiss['file_name', 'vec'](findex='./index.bin')

and this time it works!

I also run the code directly in my notebook

from towhee.functional.mixins.faiss import FaissMixin

and nothing happened.

So I guess there may be something wrong with my conda env. I will check it out later by myself.
Thanks again for your reply!