Image Segmentation with U-net and Keras!
An updated version of the code repo is available at • How to Create a Successful Github Repo for...
This is the Week 4 Deliverable for Build your own research internship in AI!
Here we look at the impact of image dimensions to data augmentation and subsequent image segmentation using the U-net and Keras.
We observe that the same model and parameters yield very different outcomes when compared to grayscale and color images.
While grayscale images lead to more false positives, colored images yield more false negatives.
This grayscale images are useful for fine vessel detection such as neo-vascularization while color images are useful for early diabetic retinopathy detection.
Google Drive: https://drive.google.com/drive/u/2/fo...
Base code:https://github.com/zhixuhao/unet
Final Github code: https://github.com/sohiniroych/Explai...
Final Video to explain complete codebase: • A Successful AI/ML Digital Portfolio, Exam...