Our great sponsors
-
WorkOS
The modern identity platform for B2B SaaS. The APIs are flexible and easy-to-use, supporting authentication, user identity, and complex enterprise features like SSO and SCIM provisioning.
Yes, I'm familiar with FAISS and consider using it to speed up the querying. But, I'm sticking with the current solution for now, because it is easier to manage when it comes to updates and removals, and actual querying is not the most time-consuming part of the "querying" execution. Currently, most of the "querying" execution time is being attributed to loading PyTorch and the CLIP model (65-73% when searching through 73 thousand images, even larger percentage on smaller datasets): https://github.com/yurijmikhalevich/rclip/issues/7. I'm interested in addressing this first. I know that TFLite can be fast, so, maybe, it's worth "porting" the model to it.
Are you familiar with the FAISS project? Should at least speed up the search piece. For processing, I bet it wouldn't be too hard to make a script that uploads batches of photos to google drive to infer their embeddings via colab.
Related posts
- 35,000 photos, what to do?
- Apple - Fruit = X? rclip update: query combos and snapcraft, homebrew, and pypi releases
- I have like 1500 videos of me playing piano.. too many to go through manually to eliminate the inappropriate ones. Can ML help?
- Can I Create my own AI Image Bank?
- Can anybody advise open-sourced neural net model to tag/recognize photos on a harddrive?