RESOURCES 01
Open datasets and benchmarks for coral reefs and marine life. Each one links to its project page with the paper, download and citation.
International Coral Reef Symposium (ICRS) 2026
Oral · 15 minLarge-scale coral reef imagery with dense semantic labels
Browse coral reef survey images online and open any one in the semantic segmentation viewer to inspect its dense labels.
CVPR 2024
HighlightSegment any COral Image on this Planet
The first foundation model for dense coral reef segmentation. It outlines individual corals at several levels of detail, from coral versus non-coral to growth form and genus, and powers the CoralSCOP-LAT tool.
IEEE Journal of Oceanic Engineering 2024
Benchmark for Dense Coral Growth Form Segmentation in the Wild
Coral images taken in real Hong Kong conditions: low visibility, motion blur, clutter and colour distortion. Every pixel is labelled by coral growth form, which helps models generalise to reefs they have not seen.
arXiv 2023
Dataset and Benchmark for Coral Video Segmentation
The first dataset for dense coral video segmentation. Segmenting every frame, rather than sampling points, gives more reliable coral coverage estimates and better 3D reef reconstruction.
WACV 2027
An Attribute-Rich Video-Text Dataset for Motion-Aware Marine Video Understanding
Marine videos annotated with fine-grained attributes (appearance, motion, environment, lighting and camera viewpoint), captions and question-answer pairs, for testing how well vision-language models understand marine video.
ECCV 2026
Advancing Event-Centric Marine Video Understanding via Visual Tool Reasoning
Underwater video events are sparse, brief and unevenly spread. MarineEVT annotates them as question-answer pairs covering semantic, contextual, spatial, temporal and causal reasoning, with the EVT-R1 tool-reasoning baseline.
WACV 2026
OralObject Recognition and Comprehension for Archiving Marine Species
A species-level benchmark that pairs bounding boxes with instance-level, expert-verified captions for detection, image grounding and captioning of marine life.
WACV 2026
OralAssessing the Marine Intelligence of Vision-Language Models
A visual question-answering benchmark that tests whether general vision-language models handle marine tasks, from species identification to conservation knowledge. Results are reported for 17 open and closed-source models.
ACM Multimedia 2025 · Datasets
A Marine Wildlife Video Dataset with Grounded Segmentation and Clip-Level Captioning
Marine wildlife videos annotated as video, caption and segmentation-mask triplets for visual grounding, captioning and video generation. Splitting videos at scene changes captures more of what each clip shows.
ECCV 2024
OralA Foundation Model for Marine Image Analysis with Instance Visual Description
A marine foundation model that outputs an instance mask and a caption for each object, trained on MarineInst20M, the largest marine image dataset to date.
MMM 2023
OralMarine Video Kit: A New Marine Video Dataset for Content-Based Analysis and Retrieval
Single-shot underwater videos from moving cameras, with metadata and keyframe descriptions for video retrieval. Used in a special session of the Video Browser Showdown 2023.