Skip to content

Latest commit

 

History

History
50 lines (38 loc) · 1.48 KB

README.md

File metadata and controls

50 lines (38 loc) · 1.48 KB
Raw

Backend

ClipSeek's backend is the core service that integrates the entire system. It utilizes FastAPI for the web server, Milvus for efficient search, PyTorch for model inference, and FFmpeg for video processing.

API Resources

ClipSeek's API follows REST principles and offers three main resources:

  1. Search
  • POST /search/by_text: Search by text.
  • POST /search/by_file: Search by file upload.
  • POST /search/by_reference: Search by indexed reference.
  • POST /search/continue: Continue an ongoing search session.
  1. Indexes
  • GET /indexes/info: Retrieve information about available indexes.
  1. File-Proxy
  • GET /resources/raw: Retrieve the original media file.
  • GET /resources/clip: Retrieve a video clip.
  • GET /resources/thumbnail: Retrieve a thumbnail from a video.

Scripts

The following scripts help you work with the system. You can run them with poetry run command.

  • compute_embeddings: Generates embeddings for different media modalities.
  • create_index: Creates Milvus collections and indexes embeddings.
  • generate_thumbnails: An optional script that generates thumbnails in advance to improve system performance.

Developing

To start developing, you have the option to use a docker container or develop locally. Following is the setup for the local environment.

Install poetry dependencies

poetry install

Run the server

make run

Run lint checkers

make lint

Prettify the code

make pretty