Glaucidius : Self-Supervised Learning on IR Imagery for Object Detection

The objective of this research is to investigate self-supervised pre-training methods that can learn good features from large unlabeled IR image and video datasets. Our goal is to significantly reduce the amount of labeled data required for the downstream task of IR image/video object detection. To do this, we will develop methods to pre-train a model by exploiting the spatial and temporal constraints of the unlabeled video data. The effectiveness of the self-supervised pre-training method will be evaluated based on its performance in downstream detection tasks.