Automatic Segmentation and Detection

Digital imaging can be a useful non-invasive tool in many biological studies. This research focuses on the challenge of manually analyzing large image sets and attempts to satisfy the need for rapid classification and measurement in changing environments using automatic segmentation and detection. This research demonstrates methods for training CNN automatic semantic segmentation algorithms using sparsely labeled data, for using transfer learning for automatic detection in large datasets and using multimodal image input (RGB + fluorescence) for segmentation. The applications of these solutions are focused on the marine environment, mostly for coral segmentation and marine animal detection.

See our automatic segmentation and detection research with coral in the press!

Nature - 2015

BBC - 2015

 

Publications

2019coralsegb.png

CoralSeg: Learning coral segmentation from sparse annotations. Iñigo Alonso, et al. Journal of Field Robotics, 2019.

2015coralsegb.png

Methods and measurement variance for field estimations of coral colony planar area using underwater photographs and semi-automated image segmentation. B. P. Neal, et al. Environmental Monitoring and Assessment, 2015.

2018marineAutomatedAnalysisc.png

Automated Analysis of Marine Video With Limited Data. Deborah Levy, et al. CVPR, 2018.

2015automatedAnnotationa.png

Towards automated annotation of benthic survey images: variability of human experts and operational modes of automation. O. Beijbom, et al. PLOS One, 2015.

2017coral-sega.png

Coral-Segmentation: Training Dense Labeling Models with Sparse Ground Truth. Iñigo Alonso, et al. ICCV, 2017.

2012doppelganger.png

Pose, Illumination and Expression Invariant Pairwise Face-Similarity Measure via Doppelganger List Comparison. Florian Schroff, et al. ICCV 2011.