CoralSeg: Learning coral segmentation from sparse annotations

Alonso, Íñigo; Yuval, Matan; Eyal, Gal; Treibitz, Tali; Murillo, Ana C.

doi:10.1002/rob.21915

Cited by 52 publications

(32 citation statements)

References 82 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…the imagery, such as temperature, irradiance, water flow, and sedimentation across the reefscape (Figure 1). As SfM is imagerybased, detailed characterization (2D/3D) of the seafloor can be undertaken through point-based annotation or semantic segmentation, with promising automation potential through machine learning (Alonso et al, 2019;Williams et al, 2019;Pavoni et al, 2020). Species-level identifications and recruit detection can be facilitated by the pairing of individual points to the original photographs (usually having greater resolution than the constructed dense point cloud), with such characterizations rapidly providing new insights into coral demographics (Edwards et al, 2017;Brito-Millán et al, 2019;Pedersen et al, 2019).…”

Section: Reefscape Characterization Through Close-range Photogrammetrymentioning

confidence: 99%

Reefscape Genomics: Leveraging Advances in 3D Imaging to Assess Fine-Scale Patterns of Genomic Variation on Coral Reefs

et al. 2021

View full text Add to dashboard Cite

Coral reefs across the world are undergoing rapid deterioration, and understanding the ecological and evolutionary processes that govern these ecosystems is critical to our ability to protect them. Molecular ecological studies have been instrumental in advancing such understanding, and while initially focused primarily on broad-scale patterns, they have gradually uncovered the prevalence of local genetic structuring. Genome-wide sequencing approaches have provided new opportunities to understand both neutral and adaptive contributions to this largely unexplained diversity, but fine-scale assessments have been hampered by challenges associated with aquatic environments, in terms of (geo)referencing, seafloor characterization, and in situ phenotyping. Here, we discuss the potential of “reefscape genomics,” leveraging recent advances in underwater imaging to enable spatially explicit genomic studies on coral reefs. More specifically, we consider how (close-range) photogrammetry approaches enable (1) fine-scale spatial mapping of benthic target organisms, (2) repeatable characterization of the abiotic and biotic reefscape, and (3) simultaneous in situ mass-phenotyping. The spatially explicit consideration of genomic data –combined with detailed environmental and phenotypic characterization– opens up the opportunity for fine-scale landscape genomic approaches on coral reefs (and other marine ecosystems). Such approaches enable assessment of the spatio-temporal drivers and adaptive potential of the extensive genetic structuring and cryptic diversity encountered in benthic invertebrates, such as reef-building corals. Considering the threats that coral reefs are facing worldwide, we believe that reefscape genomics represents a promising advancement of our molecular ecological toolkit to help inform how we can most effectively conserve and restore coral reef ecosystems into the future.

show abstract

Section: Reefscape Characterization Through Close-range Photogrammetrymentioning

confidence: 99%

Reefscape Genomics: Leveraging Advances in 3D Imaging to Assess Fine-Scale Patterns of Genomic Variation on Coral Reefs

et al. 2021

View full text Add to dashboard Cite

show abstract

“…Structural complexity can be characterized through various metrics: linear and surface rugosity (Dustan et al, 2013;Ferrari et al, 2018), fractal dimension (Tokeshi and Arakaki, 2012;Leon et al, 2015;Young et al, 2017) Broader-scale environmental parameterization has the potential to enable fine-scale modelling of further abiotic variables such as irradiance, water flow, and sedimentation across the reefscape (Figure 1). As SfM is imagery-based, detailed characterization (2D/3D) of the seafloor can be undertaken through point-based annotation or semantic segmentation, with promising automation potential through machine learning (Alonso et al, 2019;Williams et al, 2019;Pavoni et al, 2020). Species-level identifications and recruit detection can be facilitated by the pairing of individual points to the original photographs (usually having greater resolution than the constructed dense point cloud) (Edwards et al, 2017;Pedersen et al, 2019).…”

Section: Reefscape Characterization Through Close-range Photogrammetrymentioning

confidence: 99%

Reefscape genomics: leveraging advances in 3D imaging to assess fine-scale patterns of genomic variation on coral reefs

Bongaerts¹,

Dubé²,

Prata³

et al. 2021

Preprint

View full text Add to dashboard Cite

Coral reefs across the world are undergoing rapid deterioration, and understanding the ecological and evolutionary processes that govern these ecosystems is critical to our ability to protect them. Molecular ecological studies have been instrumental in advancing such understanding, and while initially focused primarily on broad-scale patterns, they have gradually uncovered the prevalence of local genetic structuring. Genome-wide sequencing approaches have provided new opportunities to understand both neutral and adaptive contributions to this largely unexplained diversity, but fine-scale assessments have been hampered by challenges associated with aquatic environments, such as (geo)referencing, seafloor characterization, and in situ phenotyping. Here, we discuss the potential of “reefscape genomics”, leveraging recent advances in underwater imaging to enable spatially-explicit genomic studies on coral reefs. More specifically, we consider how (close-range) photogrammetry approaches enable (1) fine-scale spatial mapping of benthic target organisms, (2) repeatable characterization of the abiotic and biotic reefscape, and (3) simultaneous in situ mass-phenotyping. The spatially-explicit consideration of genomic data –combined with detailed environmental and phenotypic characterization– opens up the opportunity for fine-scale landscape genomic approaches on coral reefs (and other marine ecosystems). Such approaches enable assessment of the spatio-temporal drivers and adaptive potential of the extensive genetic structuring and cryptic diversity encountered in benthic invertebrates, such as reef-building corals. Considering the threats that coral reefs are facing worldwide, we believe that reefscape genomics represents a promising advancement of our molecular ecological toolkit to help inform how we can most effectively conserve and restore coral reef ecosystems into the future.

show abstract

“…The resulting labeled masks are then used to fine-tune a SegNet [9] network. More recently, the same authors released the first extensive dataset of mask-labeled benthic images [10], obtained by the propagation of sparse annotation with a multi-level superpixel method. A custom annotation tool for the creation of segmented orthos of the seafloor, based on SLIC and graph-cut, is described in King [11].…”

Section: The Semantic Segmentation Of Coral Reefsmentioning

confidence: 99%

On Improving the Training of Models for the Semantic Segmentation of Benthic Communities from Orthographic Imagery

et al. 2020

View full text Add to dashboard Cite

The semantic segmentation of underwater imagery is an important step in the ecological analysis of coral habitats. To date, scientists produce fine-scale area annotations manually, an exceptionally time-consuming task that could be efficiently automatized by modern CNNs. This paper extends our previous work presented at the 3DUW’19 conference, outlining the workflow for the automated annotation of imagery from the first step of dataset preparation, to the last step of prediction reassembly. In particular, we propose an ecologically inspired strategy for an efficient dataset partition, an over-sampling methodology targeted on ortho-imagery, and a score fusion strategy. We also investigate the use of different loss functions in the optimization of a Deeplab V3+ model, to mitigate the class-imbalance problem and improve prediction accuracy on coral instance boundaries. The experimental results demonstrate the effectiveness of the ecologically inspired split in improving model performance, and quantify the advantages and limitations of the proposed over-sampling strategy. The extensive comparison of the loss functions gives numerous insights on the segmentation task; the Focal Tversky, typically used in the context of medical imaging (but not in remote sensing), results in the most convenient choice. By improving the accuracy of automated ortho image processing, the results presented here promise to meet the fundamental challenge of increasing the spatial and temporal scale of coral reef research, allowing researchers greater predictive ability to better manage coral reef resilience in the context of a changing environment.

show abstract

CoralSeg: Learning coral segmentation from sparse annotations

Cited by 52 publications

References 82 publications

Reefscape Genomics: Leveraging Advances in 3D Imaging to Assess Fine-Scale Patterns of Genomic Variation on Coral Reefs

Reefscape Genomics: Leveraging Advances in 3D Imaging to Assess Fine-Scale Patterns of Genomic Variation on Coral Reefs

Reefscape genomics: leveraging advances in 3D imaging to assess fine-scale patterns of genomic variation on coral reefs

On Improving the Training of Models for the Semantic Segmentation of Benthic Communities from Orthographic Imagery

Contact Info

Product

Resources

About