A CNN based image segmentation tool oriented to marine data analysis
TagLab
TagLab was created to support the activity of annotation and extraction of statistical data from ortho-maps of benthic communities. The tool includes different types of CNN-based segmentation networks specially trained for agnostic (relative only to contours) or semantic (also related to species) recognition of corals.
Interaction
TagLab allows to :
- zoom and navigate a large map using (zoom/mouse wheel, pan/’Move’ tool selected + left button). With every other tool selected the pan is activated with ctrl + left button
- segment coral instances in a semi-automatic way by simply clicks at the corals’ extremes. This is achieved using the Deep Extreme Cut network fine-tuned on coral images. Deep Extreme Cut original code : https://github.com/scaelles/DEXTR-PyTorch/
- assign a class with the ‘Assign class’ tool. Area and perimeter are now