Zero-Shot Object Detection
Definition
Computer vision models capable of localizing and classifying objects of novel, arbitrary visual categories not seen during dedicated supervised training.
Key Characteristics
- Grounding visual region proposals against text embeddings (CLIP / SigLIP).
- Open-vocabulary inference without retraining or fine-tuning.
- Real-time edge performance via model distillation (DART).
Applications
- Autonomous robotics
- Surveillance & security
- Interactive installations
Mentions in Source
- βTraining-free framework that converts SAM3 into a real-time multi-class open-vocabulary detector.β β dart-detect-anything-in-real-time