35 lines
978 B
Plaintext
35 lines
978 B
Plaintext
# Sanad Package 3 — Recognition + Places + Memories — self-contained deps.
|
|
# Comms + mask (like P2) + camera vision (opencv). Recognition is Gemini-side
|
|
# (in-context primers) — NO local ML / torch / face-embeddings.
|
|
|
|
# Web dashboard
|
|
fastapi
|
|
uvicorn[standard]
|
|
pydantic
|
|
python-multipart
|
|
websockets
|
|
|
|
# IPC bus shim + offline license verification
|
|
pyzmq
|
|
cryptography
|
|
|
|
# Numerics + image
|
|
numpy
|
|
Pillow
|
|
|
|
# Gemini Live voice + audio I/O
|
|
google-genai>=1.0.0
|
|
pyaudio
|
|
soundfile
|
|
requests
|
|
|
|
# Camera vision — JPEG-encode + USB (/dev/video*) capture. Headless (no GUI/libGL).
|
|
# pyrealsense2 is deliberately OMITTED (PyPI wheel's newer glibc ImportErrors on
|
|
# slim-bookworm); camera.py probes pyrealsense2 then falls back to cv2 V4L capture,
|
|
# so a USB colour camera works via opencv alone. Add pyrealsense2 (built from
|
|
# source) only if RealSense-specific depth is required.
|
|
opencv-python-headless
|
|
|
|
# Mask (BLE LED "Shining Mask") — expressions on recognition.
|
|
bleak==0.22.3
|