video_intelligence/config.py

"""
Video Intelligence Platform — Configuration
"""
import os
from dataclasses import dataclass, field
from typing import Optional


@dataclass
class Config:
    # ── Gemini API ──────────────────────────────────────────────
    gemini_api_key: str = field(default_factory=lambda: os.environ.get("GEMINI_API_KEY", ""))
    gemini_vision_model: str = "gemini-2.0-flash"
    gemini_embedding_model: str = "text-embedding-004"
    gemini_embedding_dim: int = 768

    # ── SigLIP2 for frame embeddings ────────────────────────────
    siglip_model: str = "google/siglip2-so400m-patch14-384"
    siglip_embedding_dim: int = 1152

    # ── Grounding DINO for attribute detection ──────────────────
    grounding_dino_model: str = "IDEA-Research/grounding-dino-tiny"
    detection_box_threshold: float = 0.35
    detection_text_threshold: float = 0.25

    # ── Frame extraction ────────────────────────────────────────
    extract_fps: float = 1.0          # frames per second to extract
    max_frames: int = 3600            # cap at 1hr @ 1fps

    # ── Indexing ────────────────────────────────────────────────
    db_path: str = "video_index.db"   # SQLite path
    faiss_visual_path: str = "visual_index.faiss"
    faiss_caption_path: str = "caption_index.faiss"

    # ── Search ──────────────────────────────────────────────────
    top_k: int = 20
    akinator_threshold: int = 10      # if results > this, start tree refinement

    # ── Device ──────────────────────────────────────────────────
    device: str = "cpu"

    def validate(self):
        if not self.gemini_api_key:
            raise ValueError(
                "GEMINI_API_KEY not set. Export it: export GEMINI_API_KEY='your-key'"
            )