Computer vision systems, such as object detection, traditionally rely on supervised learning and predetermined categories, an approach facing limitations when applied to infrared images due to dataset constraints. Emerging contrastive vision-language models, like (Contrastive Language-Image Pre-Training) CLIP, offer a transformative approach through their pre-training on extensive image-text pairs, providing diverse visual representations integrated with language semantics.
Our work proposes a novel zero-shot object detection approach for infrared images by extending the benefits of CLIP into this domain. We have developed a two-stage detection system using CLIP for detecting humans in infrared images. The first stage involves region proposal by a (You Only Look Once) YOLO object detector, followed by CLIP in the second stage. When compared with a YOLO model fine-tuned using infrared images, our proposed system demonstrates comparable performance, illustrating its efficacy as a zero-shot object detection approach. This method opens up new avenues for infrared image processing leveraging the capabilities of foundation models.
KEYWORDS: Infrared sensors, Deep learning, Quantization, Neural networks, Histograms, Embedded systems, Systems modeling, Infrared imaging, Visualization, Education and training
In recent years, the market for infrared (IR) sensors has expanded from traditional defense and security applications to include consumer products. As a result, there is increasing demand for embedded IR systems that integrate low-cost sensors with embedded processors. Meanwhile, deep learning has made significant advances and achieved superhuman performance in some domains. To support deep learning in embedded systems, new processors have emerged that are specifically designed for running deep neural networks. In this paper, we propose a performance evaluation method for applying quantized deep learning to low-cost IR sensors using layer-wise relevance propagation. Our method provides visualized analysis of what the neural networks learn. We demonstrate the effectiveness of our approach through experiments on a low-cost IR sensor dataset, showing that our method achieves an explainable performance evaluation method for degraded cases arising from the tradeoff between speed and accuracy of quantized detectors which is a typical problem in embedded systems with limited computational resources.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.