CoralVOS: Dataset and Benchmark for Coral Video Segmentation

Benchmark (surveying)
DOI: 10.48550/arxiv.2310.01946 Publication Date: 2023-01-01
ABSTRACT
Coral reefs formulate the most valuable and productive marine ecosystems, providing habitat for many species. reef surveying analysis are currently confined to coral experts who invest substantial effort in generating comprehensive dependable reports (\emph{e.g.}, coverage, population, spatial distribution, \textit{etc}), from collected survey data. However, performing dense based on manual efforts is significantly time-consuming, existing algorithms compromise opt down-sampling only conducting sparse point-based within selected frames. such will \textbf{inevitable} introduce estimation bias or even lead wrong results. To address this issue, we propose perform \textbf{dense video segmentation}, with no involved. Through object segmentation, could generate more \textit{reliable} \textit{in-depth} than algorithms. boost analysis, a large-scale segmentation dataset: \textbf{CoralVOS} as demonstrated Fig. 1. best of our knowledge, CoralVOS first dataset benchmark supporting segmentation. We experiments dataset, including 6 recent state-of-the-art (VOS) fine-tuned these VOS achieved observable performance improvement. The results show that there still great potential further promoting accuracy. trained models be released acceptance work foster research community.
SUPPLEMENTAL MATERIAL
Coming soon ....
REFERENCES ()
CITATIONS ()
EXTERNAL LINKS
PlumX Metrics
RECOMMENDATIONS
FAIR ASSESSMENT
Coming soon ....
JUPYTER LAB
Coming soon ....