SISP: A Benchmark Dataset for Fine-grained Ship Instance Segmentation in Panchromatic Satellite Images

TL;DR

The paper presents SISP, a large-scale, fine-grained ship instance segmentation dataset composed of 56,693 ship instances across 10,000 0.5 m pan-sharp satellite images, with four categories and pixel-level polygon annotations to reflect real-world scenes. It also introduces DFRInst, a benchmark network that enhances feature representation through Dynamic Feature Refinement (DFR), a DFR-assisted Feature Pyramid Network (DFR-FPN), and a DFR-assisted mask head (DFR-MH) built on a Mask R-CNN framework with a Swin Transformer backbone. Extensive experiments compare DFRInst to state-of-the-art two-stage, one-stage, and attention-based methods on SISP, showing improved performance especially for small and densely packed targets and establishing baselines for future work. The dataset and code availability aim to accelerate research on practical, fine-grained ship segmentation in satellite imagery and to support maritime monitoring applications.

Abstract

Fine-grained ship instance segmentation in satellite images holds considerable significance for monitoring maritime activities at sea. However, existing datasets often suffer from the scarcity of fine-grained information or pixel-wise localization annotations, as well as the insufficient image diversity and variations, thus limiting the research of this task. To this end, we propose a benchmark dataset for fine-grained Ship Instance Segmentation in Panchromatic satellite images, namely SISP, which contains 56,693 well-annotated ship instances with four fine-grained categories across 10,000 sliced images, and all the images are collected from SuperView-1 satellite with the resolution of 0.5m. Targets in the proposed SISP dataset have characteristics that are consistent with real satellite scenes, such as high class imbalance, various scenes, large variations in target densities and scales, and high inter-class similarity and intra-class diversity, all of which make the SISP dataset more suitable for real-world applications. In addition, we introduce a Dynamic Feature Refinement-assist Instance segmentation network, namely DFRInst, as the benchmark method for ship instance segmentation in satellite images, which can fortify the explicit representation of crucial features, thus improving the performance of ship instance segmentation. Experiments and analysis are performed on the proposed SISP dataset to evaluate the benchmark method and several state-of-the-art methods to establish baselines for facilitating future research. The proposed dataset and source codes will be available at: https://github.com/Justlovesmile/SISP.

Figures (9)

• Figure 1: Illustration of typical examples taken from the proposed SISP dataset, where the pixel-wise localization annotations for each category are displayed in the specific colour for better visualization.
• Figure 2: Visualization of the collected panchromatic satellite images for the SISP dataset, where (a), (b), (c), (d) and (e) cover different scenarios of the coastal city, island, offshore, lake and river, respectively.
• Figure 3: Illustration of typical examples from the SISP dataset annotated in different formats, i.e., polygon, HBB and OBB.
• Figure 4: Statistics of category distribution in the SISP dataset. (a) Histogram of the number of instances per category. (b) Histogram of the number of categories per image.
• Figure 5: Statistics of instances and densities in the SISP dataset. (a) Histogram of the number of instances per image. (b) Histogram of the density distribution for each category. Here, density is calculated by the closet distance between two targets (i.e., polygons), where dense means the distance is less 10 pixels, normal indicates the distance is between 10 and 50 pixels, and sparse means the distance is more than 50 pixels.