docs/known-objects.md

Known Object Detection (RT-DETR)

Overview

Pipeline 1 leverages RT-DETR (Real-Time Detection Transformer) to deliver fast, highly accurate bounding box predictions for standard object categories.

Image Frame ──► RT-DETR Backbone ──► Known Class Bounding Boxes + Confidence

Key Characteristics

• Predictable Performance: Extremely consistent bounding boxes and confidence scores.
• Fixed Vocabulary: Trained on COCO 80 categories (person, car, dog, bottle, chair, laptop, etc.).
• Ideal for Real-Time Feeds: Runs efficiently on both GPU and modern CPU setups.

Code Example

from object_intelligence import ObjectDetector
import cv2

detector = ObjectDetector()
detector.set_mode("known")

cap = cv2.VideoCapture(0)
while True:
    ret, frame = cap.read()
    if not ret: break
    annotated, detections = detector.detect_and_draw(frame)
    cv2.imshow("Known Objects", annotated)
    if cv2.waitKey(1) & 0xFF == ord('q'): break
cap.release()