Marine Object Detection using Transformers

Underwater object detection is a challenging task due to low visibility and complex background scenes. Although different CNN architectures have been proposed in the literature for this task but they lack the ability to utilize global contextual information about the image. Visual transformer-based architectures have recently gained attention after the success of transformers in NLP. In this paper, DEtection TRansformer (DETR), a visual transformer-based object detection architecture, is utilized for the underwater object detection task. The fine-tuning of the model is carried out on the Brackish dataset with 6 different classes. Exploiting the global contextual information and relationship of objects in the scene reports promising performance on standard evaluation metrics.