Object-guided contrastive language-image pre-training for zero-shot target recognition

Key Points

• Zero-shot target recognition is achieved with a notable accuracy of 85% in benchmark tests.
• Contrastive learning effectively links images and language representations, enhancing recognition tasks.
• The analysis employs a novel object-guided approach to improve learning efficiency and relevance.
• These findings support the practicality of pre-training models for real-world deployment in various domains.