VideoSeek

中文说明 | English

Desktop semantic video search built with PySide6, ONNX Runtime, FAISS, and FFmpeg.

1. Install dependencies:

pip install -r requirements.txt

Or install explicitly (Windows; includes mobile bridge and QR):

pip install onnxruntime-directml opencv-python PySide6 faiss-cpu numpy pillow tokenizers ftfy regex yt-dlp python-vlc fastapi uvicorn python-multipart "qrcode[pil]"

On Linux or macOS, use onnxruntime instead of onnxruntime-directml (this project’s GPU path targets Windows DirectML; inference falls back to CPU elsewhere).

2. Start the app:

python main.py

3. First launch — models and FFmpeg are not bundled. Recommended: download the official zip from 123 cloud drive (models), run python main.py, open Import runtime resources, add the .zip, then Import and Parse (details in docs/quickstart.md § 3.1). Advanced manual layout is in § 3.2.

• Model files depend on the active model profile (default clip_onnx). Primary workflow: get the maintainer-built zip from 123 cloud drive (models) and import it in the app (same steps as quickstart § 3.1). src/app/app_meta.py still holds URLs for release notes / version / about and may point 「Go to download」 at the cloud folder; it is not assumed that ONNX files are fetched automatically without that zip flow.
• Default clip_onnx example files:
  • clip_visual.onnx
  • clip_text.onnx
  • bpe_simple_vocab_16e6.txt.gz
• ffmpeg.exe for frame extraction and preview.
• python-vlc plus VLC runtime for in-app preview playback.

Windows VLC runtime layout (project-local):

• vlc_lib/libvlc.dll
• vlc_lib/libvlccore.dll
• vlc_lib/plugins/

If vlc_lib/ is missing or incomplete, search/indexing may still work but preview playback may be unavailable.

• Search local video libraries with text or image.
• Build and search a remote library from online links.
• Preview matched clips inside the app.

Focused subset:

python -m unittest ^
  tests.test_runtime_resource_service ^
  tests.test_notice_version_utils ^
  tests.test_download_services ^
  tests.test_controllers

Full suite: python -m unittest discover -s tests -p "test_*.py" (see docs/quickstart.md).

• Detailed setup and troubleshooting: docs/quickstart.md
• Architecture and module map: docs/architecture.md

MIT