Coral-Segmentation: Training Dense Labeling Models with Sparse Ground Truth

Alonso, Íñigo; Cambra, Ana B.; Muñoz, Arsenio; Treibitz, Tali; Murillo, Ana C.

doi:10.1109/iccvw.2017.339

Cited by 31 publications

(45 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In our case where only some sparse labels are available, there are two existing approaches for addressing the sparsity: either propagate the sparse labels into dense labels or, train only on the sparse labels and ignore the nonlabeled pixels. We previously showed (Alonso et al, ) that the first approach provides better results as it provides more data for training.…”

Section: Training Dense Semantic Segmentation With Sparse Pixel Labelsmentioning

confidence: 99%

“…Initially, we consider a simple but intuitive approach: single‐level superpixel‐based augmentation. This strategy, detailed in our preliminary work (Alonso et al, ), takes an input image with sparse labels and augments them in two steps. First, the image is segmented into a preset number of superpixels, as shown in the examples in Figure .…”

Section: Training Dense Semantic Segmentation With Sparse Pixel Labelsmentioning

confidence: 99%

“…This single‐level superpixel strategy has been used in prior works with promising results (Alonso et al, ; King et al, ) but has some drawbacks because the number of superpixels has to be specified a priori. Consequently, two issues can potentially arise.…”

Section: Training Dense Semantic Segmentation With Sparse Pixel Labelsmentioning

confidence: 99%

“…This is an approximated labeling because it contains some noise. It does, however, provide a very representative reference labeling (Alonso & Murillo, ; Alonso et al, ). Dense‐GT : We use this only for the Eilat data set. It contains a few dense labeled images for binary (coral vs. noncoral) segmentation obtained by an expert coral biologist.…”

Section: Data Sets and Labelsmentioning

confidence: 99%

“…Tables and show the comparison between the single‐level augmentation used in recent previous work (Alonso et al, ) and other work that followed aimed at building an annotation tool (King et al, ), and the proposed multilevel augmentation (ours) using different superpixel segmentation techniques.…”

Section: Evaluation Of Our Proposed Approachmentioning

confidence: 99%

See 4 more Smart Citations

CoralSeg: Learning coral segmentation from sparse annotations

Alonso

Yuval

Eyal

et al. 2019

Journal of Field Robotics

Self Cite

View full text Add to dashboard Cite

Robotic advances and developments in sensors and acquisition systems facilitate the collection of survey data in remote and challenging scenarios. Semantic segmentation, which attempts to provide per‐pixel semantic labels, is an essential task when processing such data. Recent advances in deep learning approaches have boosted this task's performance. Unfortunately, these methods need large amounts of labeled data, which is usually a challenge in many domains. In many environmental monitoring instances, such as the coral reef example studied here, data labeling demands expert knowledge and is costly. Therefore, many data sets often present scarce and sparse image annotations or remain untouched in image libraries. This study proposes and validates an effective approach for learning semantic segmentation models from sparsely labeled data. Based on augmenting sparse annotations with the proposed adaptive superpixel segmentation propagation, we obtain similar results as if training with dense annotations, significantly reducing the labeling effort. We perform an in‐depth analysis of our labeling augmentation method as well as of different neural network architectures and loss functions for semantic segmentation. We demonstrate the effectiveness of our approach on publicly available data sets of different real domains, with the emphasis on underwater scenarios—specifically, coral reef semantic segmentation. We release new labeled data as well as an encoder trained on half a million coral reef images, which is shown to facilitate the generalization to new coral scenarios.

show abstract

Section: Training Dense Semantic Segmentation With Sparse Pixel Labelsmentioning

confidence: 99%

Section: Training Dense Semantic Segmentation With Sparse Pixel Labelsmentioning

confidence: 99%

Section: Training Dense Semantic Segmentation With Sparse Pixel Labelsmentioning

confidence: 99%

Section: Data Sets and Labelsmentioning

confidence: 99%

Section: Evaluation Of Our Proposed Approachmentioning

confidence: 99%

See 3 more Smart Citations