ABOShips -- An Inshore and Offshore Maritime Vessel Detection Dataset with Precise Annotations

TL;DR

This paper introduces ABOShips, a comprehensive maritime vessel dataset with precise annotations, and evaluates the performance of four popular object detection algorithms on this dataset, highlighting their strengths and weaknesses.

Contribution

The paper provides a new, detailed maritime vessel dataset with rigorous annotations and benchmarks four detection algorithms, revealing their relative performance in inshore and offshore conditions.

Findings

Faster R-CNN with Inception-Resnet v2 performs best overall.

EfficientDet surpasses Faster R-CNN in large object detection.

The dataset includes 9 vessel types, seamarks, and floaters with varied conditions.

Abstract

Availability of domain-specific datasets is an essential problem in object detection. Maritime vessel detection of inshore and offshore datasets is no exception, there is a limited number of studies addressing this need. For that reason, we collected a dataset of images of maritime vessels taking into account different factors: background variation, atmospheric conditions, illumination, visible proportion, occlusion and scale variation. Vessel instances (including 9 types of vessels), seamarks and miscellaneous floaters were precisely annotated: we employed a first round of labelling and subsequently, we used the CSRT [1] tracker to trace inconsistencies and relabel inadequate label instances. Moreover, we evaluated the the out-of-the-box performance of four prevalent object detection algorithms (Faster R-CNN [2], R-FCN [3], SSD [4] and EfficientDet [5]). The algorithms were previously trained on the Microsoft COCO dataset. We compare their accuracy based on feature extractor and object size. Our experiments show that Faster R-CNN with Inception-Resnet v2 outperforms the other algorithms, except in the large object category where EfficientDet surpasses the latter.