VesselBench-800K: A Large-scale Perception Benchmark for Multimodal Vessel Detection, Counting, and Density Estimation

Danfeng Hong; Chenyu Li; Jocelyn Chanussot

cs.CVSep 29, 2026

VesselBench-800K: A Large-scale Perception Benchmark for Multimodal Vessel Detection, Counting, and Density Estimation

Authors: Danfeng Hong, Chenyu Li, Jocelyn Chanussot

Organizations: School of Automation, Southeast University, Nanjing 211189, China · Univ. Grenoble Alpes, INRAE, LESSEM, 38000 Grenoble, France · Univ. Grenoble Alpes, Inria, CNRS, Grenoble INP, LJK, Grenoble 38000, France

Abstract

Vessel perception from space is crucial for a wide range of maritime applications, from traffic monitoring to environmental protection. However, most existing datasets predominantly focus on general object detection tasks in optical remote sensing (RS) images. Relying solely on single-modality optical RS images proves inadequate for effectively perceiving vessel objects in complex maritime scenarios, where ever-changing weather conditions (e.g., clouds and rain), the need for day-and-night coverage, and the inherent limitations of a single imaging modality pose significant challenges. To fill this gap, we introduce VesselBench-800K, the largest-to-date benchmark dataset on a global scale for vessel perception in multimodal RS images. As its name suggests, VesselBench-800K comprises 800,000 images, each at a resolution of 512x512 pixels, specifically curated for vessel perception tasks such as detection, counting, and density estimation. These multimodal image pairs (i.e., optical, SAR) are collected from diverse platforms, sensors, scenes, shooting heights, and synthetic sources, spanning spatial resolutions from 4.5m to 0.1m. Furthermore, we evaluate numerous state-of-the-art detection, counting, and density estimation models on VesselBench-800K through both qualitative and quantitative comparisons. By revealing previously unrecognized cues, this dataset holds immense potential to significantly advance our understanding of marine traffic. Our VesselBench dataset will be publicly available at https://github.com/danfenghong/IEEE_TGRS_VesselBench to support and contribute to community development.

Figures & tables

Fig. 1: We introduce VesselBench-800K , a large-scale multimodal (passive Optical and active SAR ) remote sensing (RS) dataset for vessel perception, collected globally. VesselBench-800K features 800,000 images of 512 $\times$ 512 pixels, with ground sampling distance (GSD) ranging from 4.5m to 0.1m. Vessels are labeled in a horizontal bounding box (HBB) fashion, highlighted in green , and ground truth (GT) maps for density estimation are generated by means of Gaussian kernel functions. VesselBench-800K caters to three main tasks: detection, counting, and density estimation, making it a comprehensive tool for vessel perception studies.

Datasets	# of images	Image size (px.)	GSD (m)	# Instances	Annot.	Div.	Tasks	Modality	Data sources
Optical : Vessel subset in multi-class object detection datasets
LEVIR [ 11 ]	22,000	800 $\times$ 600	0.2 $\sim$ 1.0	3,025	HBB	✔	Det	RGB	GE (Google Earth)
DIOR [ 12 ]	23,463	800 $\times$ 800	0.5 $\sim$ 30	62,400	OBB	✔	Det	RGB	GE
NWPU VHR-10 [ 13 ]	800	497 $\times$ 693 $\sim$ 606 $\times$ 1,100	0.08 $\sim$ 2.0	–	HBB	✘	Det	RGB	GE, Vaihingen
DOTA-v1.0 [ 5 ]	2,806	800 $\times$ 800 $\sim$ 4,000 $\times$ 4,000	0.5	20,869	OBB	✘	Det	RGB	GE, JL-1, GF-2, Aerial
DOTA-v2.0 [ 14 ]	11,268	800 $\times$ 800 $\sim$ 20,000 $\times$ 20,000	0.5	42,517	OBB	✘	Det	RGB	GE, JL-1, GF-2, Aerial

TABLE I: Quantitative comparisons between our VesselBench-800K and existing vessel detection/perception datasets from unimodal (optical, SAR) and multimodal perspectives in terms of nine indices: the number of images (# of images), image size (pixels), GSD (resolution), number of vessels (# instances), types of annotations (Annot.), diverse backgrounds (Div.), types of tasks (Det: detection, C: counting, E: density estimation), modality, and data sources, respectively.

Fig. 2: Typical examples from VesselBench-800K to illustrate key data characteristics, including diverse vessel types, sizes, and densities, varied scenarios (ranging from simple to complex), and multi-scale vessels within the same and different scenes.

Sources	# of images	Modality	GSD (m)	Labels
DOTA	$\sim$ 22,000	RGB	0.5	re-w/
GLH-water	$\sim$ 400	RGB	0.3	w/o
GLH-Bridge	$\sim$ 80,000	RGB	0.3 $\sim$ 1.0	w/o
QXS-SAROPT	$\sim$ 230	RGB, SAR	1.0	w/o
New data	$\sim$ 697,370	RGB, SAR	0.1 $\sim$ 4.5	w/o

TABLE II: Detailed composition of VesselBench-800K : the number of images (of # images), involved modalities, GSD, and labeling status (re-w/: re-labeled, w/o: without labels).

Fig. 3: An illustrative data statistics of VesselBench-800K. (a) Spatial distribution of data collection on a global scale, categorized into training, validation, and testing sets in red , green , and blue , respectively. (b) gives the number of images across different resolutions (GSD). (c) shows pixel-level statistics of vessel instances. (d) collects the number of HBB per image. (e) counts the number of instances in different aspect ratios of HBB. (f) calculates the area ratio of targets within images.

Unimodal Optical vessel detection
Methods	Backbone	Epochs	AP	AP 50	AP 75	AP s	AP m	AP l	Parameters	FLOPs
Faster R-CNN [ 60 ]	ResNet50-FPN	36	0.512	0.761	0.568	0.400	0.731	0.707	41.348M	0.148T
RetinaNet [ 61 ]	ResNet50-FPN	36	0.536	0.762	0.573	0.402	0.750	0.752	36.351M	0.128T
Mask R-CNN [ 62 ]	ResNet50-FPN	36	0.552	0.736	0.617	0.405	0.785	0.831	43.971M	0.276T
AutoAssign [ 63 ]	ResNet50-FPN	36	0.532	0.752	0.591	0.399	0.742	0.712	36.426M	0.126T
TOOD [ 64 ]	ResNet50-FPN	36	0.586	0.775	0.641	0.445	0.807	0.822	32.018M	0.123T

TABLE III: Quantitative performance comparison of vessel detection methods using unimodal and multimodal RS images on VesselBench-800K, where eight metrics, including mAP, AP 50 , AP 75 , AP s , AP m , AP l , as well as FLOPs and parameters of networks, are compared. FPN refers to feature pyramid networks [ 58 ] and CM denotes ChannelMapper [ 59 ] . The best results for different modalities are highlighted in bold, respectively.

Fig. 4: Visualization of exemplar results from different vessel detection methods in unimodal and multimodal scenarios on VesselBench-800K. The HBBs in green, red, and yellow denote true positives, false negatives, and false positives, respectively.

Fig. 5: Visualization of different compared methods for vessel counting and density estimation in the selected five scenes of VesselBench-800K. Note that the heat maps reflect the density estimation for vessels, while the numbers in red denote the GT and estimated results of vessel counting.

Methods	Epochs	MAE	MSE	PSNR	SSIM
MCNN [ 67 ]	27	2.714	35.468	23.088	0.764
CSRNet [ 68 ]	26	1.524	20.277	29.382	0.926
SCAR [ 69 ]	18	1.417	20.860	31.610	0.958
SFCN+ [ 70 ]	26	1.462	19.726	31.193	0.945
PET [ 71 ]	200	1.629	3.98	–	–

TABLE IV: Quantitative results of different compared methods for vessel density estimation using multimodal RS images on VesselBench-800K, where four main indices, i.e., MAE, MSE, PSNR, and SSIM, are computed. The best results are highlighted in bold.

Explore similar work

Jul 20, 2026cs.CV

SAR Vessel Detection and Gross Tonnage Estimation from Heterogeneous Datasets for Dark Vessel Identification

Detecting vessels engaging in illegal activities is of paramount importance for maritime security. One of the major goals is to detect dark vessels, ships that disable their transponders to evade surveillance. Deep Learning (DL) models can detect vessels in Synthetic Aperture Radar (SAR) images, enabling maritime traffic analysis regardless of weather or visibility conditions. However, to detect potential dark vessels, a DL model must select only those that are required to carry a transponder based on their Gross Tonnage (GT). Unfortunately, no public SAR dataset is available for training an end-to-end DL model for vessel detection and GT regression. In this work, we present a framework that leverages heterogeneous image and tabular datasets to solve this task. Our solution combines a multi-task DL framework for predicting the location, vessel type, and physical dimensions of ships, cascaded with a non-parametric model for predicting GT from vessel size and category. We perform GT regression by a KNN that measures sample similarity using a hybrid Euclidean and categorical distance. Experiments show that our solution can predict multiple outputs while remaining competitive with state-of-the-art models on individual subtasks, thus enabling the identification of dark vessels. We publish our code on GitHub https://github.com/PaltrinieriDavide/vesseldetection.

Davide Paltrinieri, Andrea Diecidue, Roberto Basla +3

Politecnico di Milano, Milano, Italy

Ship Detection Unsupervised Detection Maritime Navigation Synthetic Aperture Radar Deep Learning

Ask PDF

Aug 30, 2026cs.CV

MariSat: A Maritime Dataset for Instance Segmentation of Objects in Satellite and Aerial Images

Automated maritime surveillance from satellite and aerial imagery requires large, precisely annotated datasets, which remain scarce for the instance-segmentation task, particularly for small vessels in cluttered port environments. We present MariSat, a new benchmark dataset of 1260 aerial and satellite images covering diverse port and coastal scenes, annotated at the pixel level for eight maritime object classes (sailboat, yacht, jet-ski, fishing boat, cruise ship, military vessel, tugboat and cargo ship). The dataset was produced through a semi-automatic annotation pipeline combining the textpromptable segmentation model SAM 3 with a cascade of geometric and colorimetric post-processing filters, followed by a manual correction and quality-control pass performed with the CVAT annotation platform. We describe the image-collection methodology, the annotation and correction process, and the resulting data organization. We also report class-wise statistics for the training, validation, and test splits. MariSat has already been used to fine-tune and benchmark segmentation and detection models (SAM 3 and YOLO11) for real-time maritime monitoring. We report detailed quantitative and per-class results for both tasks. The MariSat dataset is publicly available on GitHub : https://github.com/amirabbes/P2M-Maritime-Segmentation

Amir Abbes, Ines Harrabi, Lucas Justin Yirepoa Kinda +3

University of Carthage, SUP’COM, LR11TIC05, MEDIATRON, 2083, Ariana, Tunisia · University of Gabes, Hatem Bettaher IResCoMath Laboratory, Gabes, Tunisia · University of Rouen Normandie, ESIGELEC, IRSEEM, 76000 Rouen, France

Ship Detection Vision Datasets Aerial Imagery Maestro Dataset Objects

Ask PDF

Aug 3, 2026cs.CV

OSSDD - a New Open Dataset for Sentinel-1 Ship Detection

Ship detection in Synthetic Aperture Radar (SAR) images plays an important role for maritime situational awareness, especially with respect to different illegal activities at sea such as illegal fishing, smuggling or border violations. Modern ship detection methods using neural networks usually require large training datasets, which are considerably scarcer in the SAR domain than in the electro-optical domain. While several free datasets exist for this task, their availability and usability vary. In this paper, OpenSARShip-Ship Detection Dataset (OSSDD), a new dataset based on the well-known OpenSARShip 1.0 dataset is proposed for training neural networks for SAR ship detection. OSSDD is freely available and contains 15,197 Sentinel-1 amplitude patches in VV and VH polarization, binary ship masks, axis-aligned bounding box and rotated bounding box annotations for a total of 55,759 ships. The construction of the dataset, the contents and structure of the downloadable data and experiments with three common detector models (Faster R-CNN, FCOS, DETR) are shown and discussed. The results serve as benchmarks for future experiments. The dataset is available on Hugging Face at https://huggingface.co/datasets/sylviaHoch/OpenSARShip-Ship-Detection-Dataset.

Horst Hammer, Sylvia Hochstuhl, Antje Thiele +4

Fraunhofer Institute of Optronics, System Technologies and Image Exploitation - IOSB, Gutleuthausstraße 1, Ettlingen, 76275, Germany · Karlsruhe Institute of Technology - KIT, Kaiserstraße 12, Karlsruhe, 76131, Germany · HENSOLDT Sensors GmbH, Wörthstraße 85, Ulm, 89077, Germany

Ship Detection Synthetic Aperture Radar

Ask PDF