FHDO ImageCLEF 2020 Coral Submission Code

This repository contains the training and evaluation code used for our ImageCLEF 2020 Coral submission.

Getting started

• clone this repository
• clone Mask_RCNN
• make sure the dependencies listed below are installed
• prepare the datasets (see below)
• adjust the paths beginning with /path/to in all scripts
• run python3 prepare.py run_name followed by python3 train.py run_name

Dependencies

• OpenCV 4.2.0
• Tensorflow GPU 2.1
• numpy 1.18.2
• progressbar2 3.51.0

Dataset pre-processing

• the scripts assume the images are all resized to 1536x1536 beforehand
• for IBLA, we used this script
• for Rayleigh, we used this script on top of the IBLA images
• for color reduction, we used octree_color_quantizer

Citation

M. Arendt, J. Rückert, R. Brüngel, C. Brumann, C.M. Friedrich, "The effects of colour enhancement and IoU optimisation on object detection and segmentation of coral reef structures" Manuscript submitted for publication.