FAIR1M: A benchmark dataset for fine-grained object recognition in high-resolution remote sensing imagery

With the rapid development of deep learning, many deep learning-based approaches have made great achievements in object detection tasks. It is generally known that deep learning is a data-driven approach. Data directly impact the performance of object detectors to some extent. Although existing datasets include common objects in remote sensing images, they still have some scale, category, and image limitations. Therefore, there is a strong requirement for establishing a large-scale object detection benchmark for high-resolution remote sensing images. In this paper, we propose a novel benchmark dataset with more than 1 million instances and more than 40,000 images for Fine-grAined object recognItion in high-Resolution remote sensing imagery which is named as FAIR1M. We collected remote sensing images with a resolution of 0.3 m to 0.8 m from different platforms, which are spread across many countries and regions. All objects in the FAIR1M dataset are annotated with respect to 5 categories and 37 subcategories by oriented bounding boxes. Compared with existing detection datasets that are dedicated to object detection, the FAIR1M dataset has 4 particular characteristics: (1) it is much larger than other existing object detection datasets both in terms of the number of instances and the number of images, (2) it provides richer fine-grained category information for objects in remote sensing images, (3) it contains geographic information such as latitude, longitude and resolution attributes, and (4) it provides better image quality due to the use of a careful data cleaning procedure. Based on the FAIR1M dataset, we propose three fine-grained object detection and recognition tasks. Moreover, we evaluate several state-of-the-art approaches to establish baselines for future research. Experimental results indicate that the FAIR1M dataset effectively represents real remote sensing applications and is quite challenging for existing methods. Considering the fine-grained characteristics, we improve the evaluation metric and introduce the idea of hierarchy detection into the algorithms. We believe that the FAIR1M dataset will contribute to the earth observation community via fine-grained object detection in large-scale real-world scenes. FAIR1M Website: http://gaofen-challenge.com/.

Automated summary

With the rapid development of deep learning, many deep learning-based approaches have achieved great success in object detection tasks. However, existing datasets have limitations in terms of scale, category, and image. To address the needs of high-resolution remote sensing images, we propose a novel benchmark dataset called FAIR1M, which includes over 1 million instances and more than 40,000 images for fine-grained object recognition. The FAIR1M dataset has several unique characteristics, such as being larger than other datasets, providing richer category information, containing geographic information, and having better image quality. We evaluate state-of-the-art methods on the FAIR1M dataset and propose improvements to the evaluation metrics and the incorporation of hierarchy detection. We believe that the FAIR1M dataset will contribute to the field of earth observation through fine-grained object detection in large-scale real-world scenes.