File size: 7,692 Bytes
fea499e | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 | """
Video Intelligence Platform — Indexing Pipeline
Orchestrates the full offline indexing: frame extraction → detection → captioning → embedding → storage.
"""
import io
import time
import numpy as np
from pathlib import Path
from typing import List, Optional
from PIL import Image
from .config import Config
from .frame_extractor import extract_frames, get_video_info, ExtractedFrame
from .gemini_client import GeminiClient
from .visual_encoders import SigLIPEncoder, GroundingDINODetector
from .index_store import VideoIndex
class IndexingPipeline:
"""
Full offline indexing pipeline for a video file.
Pipeline steps:
1. Extract frames at configured FPS
2. Run Grounding DINO for attribute detection on each frame
3. Generate visual embeddings with SigLIP2
4. Caption frames with Gemini 2.0 Flash
5. Generate caption embeddings with Gemini text-embedding-004
6. Store everything in SQLite + FAISS
"""
def __init__(self, config: Config):
self.config = config
config.validate()
# Initialize components
print("=" * 60)
print("🚀 Initializing Video Intelligence Pipeline")
print("=" * 60)
self.gemini = GeminiClient(
api_key=config.gemini_api_key,
vision_model=config.gemini_vision_model,
embedding_model=config.gemini_embedding_model,
)
self.siglip = SigLIPEncoder(
model_name=config.siglip_model,
device=config.device,
)
self.detector = GroundingDINODetector(
model_name=config.grounding_dino_model,
device=config.device,
box_threshold=config.detection_box_threshold,
text_threshold=config.detection_text_threshold,
)
self.index = VideoIndex(
db_path=config.db_path,
visual_dim=config.siglip_embedding_dim,
caption_dim=config.gemini_embedding_dim,
)
print("=" * 60)
print("✅ All components initialized")
print("=" * 60)
def index_video(self, video_path: str,
caption_every_n: int = 1,
detect_every_n: int = 1) -> dict:
"""
Index a video file end-to-end.
Args:
video_path: Path to video file
caption_every_n: Caption every Nth frame (1 = all, 5 = every 5th)
detect_every_n: Run detection every Nth frame
Returns:
Dict with indexing statistics
"""
start_time = time.time()
# Step 0: Get video info and register
print("\n📊 Step 0: Registering video...")
info = get_video_info(video_path)
video_id = self.index.register_video(
path=video_path,
duration_sec=info["duration_sec"],
fps=info["fps"],
width=info["width"],
height=info["height"],
)
print(f" Video ID: {video_id} | {info['duration_sec']:.1f}s | {info['width']}x{info['height']}")
# Step 1: Extract frames
print("\n🎬 Step 1: Extracting frames...")
frames = extract_frames(
video_path,
fps=self.config.extract_fps,
max_frames=self.config.max_frames,
)
# Step 2: Store frames and run detection
print(f"\n🔍 Step 2: Running object detection on {len(frames)} frames...")
frame_ids = []
pil_images = []
for i, frame in enumerate(frames):
# Store frame in DB
frame_id = self.index.add_frame(
video_id=video_id,
frame_idx=frame.frame_idx,
timestamp_sec=frame.timestamp_sec,
)
frame_ids.append(frame_id)
pil_images.append(frame.image)
# Run detection
if i % detect_every_n == 0:
try:
detections = self.detector.detect_default_attributes(frame.image)
self.index.add_detections(frame_id, [
{"label": d.label, "confidence": d.confidence, "bbox": d.bbox}
for d in detections
])
if detections:
labels = [d.label for d in detections]
print(f" Frame {i}/{len(frames)} [{frame.timestamp_sec:.1f}s]: {labels}")
except Exception as e:
print(f" ⚠️ Detection failed for frame {i}: {e}")
if (i + 1) % 50 == 0:
print(f" ... processed {i + 1}/{len(frames)} frames")
# Step 3: Generate visual embeddings (SigLIP2)
print(f"\n🧠 Step 3: Generating visual embeddings...")
visual_embeddings = self.siglip.embed_frames(pil_images, batch_size=4)
self.index.add_visual_embeddings_batch(frame_ids, visual_embeddings)
print(f" ✅ {len(visual_embeddings)} frame embeddings stored ({visual_embeddings.shape})")
# Step 4: Caption frames with Gemini
print(f"\n📝 Step 4: Captioning frames with Gemini...")
captions = []
for i, frame in enumerate(frames):
if i % caption_every_n != 0:
captions.append("")
continue
try:
# Convert PIL to JPEG bytes
buf = io.BytesIO()
frame.image.save(buf, format="JPEG", quality=85)
jpeg_bytes = buf.getvalue()
caption = self.gemini.caption_frame(jpeg_bytes)
captions.append(caption)
self.index.update_caption(frame_ids[i], caption)
if caption:
print(f" Frame {i} [{frame.timestamp_sec:.1f}s]: {caption[:100]}...")
# Rate limiting: Gemini free tier = ~15 RPM
time.sleep(4.0)
except Exception as e:
print(f" ⚠️ Captioning failed for frame {i}: {e}")
captions.append("")
time.sleep(2.0)
# Step 5: Generate caption embeddings (Gemini text-embedding-004)
print(f"\n📐 Step 5: Generating caption embeddings...")
non_empty_captions = [(i, c) for i, c in enumerate(captions) if c]
if non_empty_captions:
caption_texts = [c for _, c in non_empty_captions]
try:
caption_embeddings = self.gemini.embed_texts(caption_texts, task_type="RETRIEVAL_DOCUMENT")
emb_frame_ids = [frame_ids[i] for i, _ in non_empty_captions]
emb_array = np.array(caption_embeddings, dtype=np.float32)
self.index.add_caption_embeddings_batch(emb_frame_ids, emb_array)
print(f" ✅ {len(caption_embeddings)} caption embeddings stored")
except Exception as e:
print(f" ⚠️ Caption embedding failed: {e}")
# Step 6: Save indices
print(f"\n💾 Step 6: Saving indices...")
self.index.save_faiss(
self.config.faiss_visual_path,
self.config.faiss_caption_path,
)
elapsed = time.time() - start_time
stats = self.index.stats()
stats["elapsed_sec"] = elapsed
print(f"\n{'=' * 60}")
print(f"✅ Indexing complete!")
print(f" Frames: {stats['frames']}")
print(f" Detections: {stats['detections']}")
print(f" Visual vectors: {stats['visual_vectors']}")
print(f" Caption vectors: {stats['caption_vectors']}")
print(f" Time: {elapsed:.1f}s")
print(f"{'=' * 60}")
return stats
|