By performing image analysis, can we detect and identify kernels (i.e., seeds) that are normal versus kernels that will become the purple kernel eater monsters? You will receive two images taken on consecutive days (i.e., one day goes between images). They all seem to start out normal but can you detect something consistently about the kernel that will clue you in on whether or not it is normal or will turn into a purple monster?
In this challenge, you are asked to propose some algorithms which, given a day 0 image (with no discernible purple in it), indicate the kernels which will become purple one day later (i.e. which are purple in the Day 1 image).
In addition, we would like to know your thoughts on what changes in our imaging style (background, lighting, sharpness, etc.) could help you develop a better algorithm - Basically data observations and thoughts around additional data gathering.
Please check the following two images as an example.
Day 0 image:
Day 1 image:
As you can see here, there might be some changes in the imaging styles (e.g., background, lighting, sharpness etc.). So you algorithm should be robust to these differences.
Full dataset: https://drive.google.com/open?id=1G1IRNQq12duHkinq3zPoClWFLS4Vdirf
Some clues to help you identify the non-purple monsters:
The purple monsters are hungry and eat a lot. Therefore, they are likely to be slightly different (Obese, fat, or shaped differently)
The purple monsters may have some strange colors even in the initial image. We don’t know much about this though.
The purple monsters may have other features which are different from others e.g. their surface may have different texture, or they may have a different aspect ratio. There may be other features which we have not found yet (but you are welcome to find and use)