So I’m working on this project involving CometAssay, which is a method to measure DNA damage in cells. I use images and try to measure the lengths of said damage. In this case, I have two classes I’m trying to predict: the comet head, which is the bright circle-looking thing, and the entire comet, which is essentially the entire cell (so it includes the head and the extra tail coming off of it). I want to specifically measure the “tail” length, which is the dash coming off of the bright circle. To do this, I’ve chosen to measure the entire comet, then subtract the length of the head from that to get the tail length, simply because it was easier to train a model this way.
I’ve gotten a prediction working for the head and the comet, but I’m trying to find a way that I can use the JSON file from the prediction to identify which element is which and where. The file tells me an x and y location of either the head class or the comet class, but it doesn’t label it on the image, so I’m stuck with trying to figure out which class is where. I’m not sure how to actually find the right classes for the head and comet class and label it so I can do the math to find the tail length. Ideally, I would have all the comets and head classes labeled so I can do some simple subtraction to find the tail length, and then annotate it back onto the predicted image.
So here’s a picture of the predicted image and json file. The top left image is the before prediction and the bottom is the after. The JSON file is on the right. Sorry I had to combine the images because I’m a new user and they won’t let me put multiple images in a post…
