Coral Image Segmentation

During the Spring of 2020, I utilized deep learning segmentation to do coral segmentation. I used a dataset of ~ 150 images of corals, which contains eight different categories.

• Pre-processed images using PIL and OpenCV and annotated. I used different techniques, including but not limited to, histogram equalization, white balance, color balance.
• I have done data augmentation because of the small size of the dataset. Methods used: cropping, rotation, mirror and etc.
• Dataset annotated carefully using labelme.
• Used different ResNet, U-Net, FCN, Vanilla CNN, and etc.
• Hyperparameter tuning.
• Segmented images of 8-different categories of corals with the accuracy of mIoU 60.1 %, considering the size of the dataset and the time I had, It’s a good one.