Hi!
I am starting on image detection (so sorry if this is kind of nooby post) in a dataset with only one category, the idea would be to detect a specific type of cells in a microscopy image like here:
I have a folder with 300x300 images and a csv with the following columns:
image, label, xmin, xmax, ymin, ymax
So I guess I have all the important requirements for the training but I am not sure how to handle a single label when using ObjectItemList. I am following the COCO tiny example, but I am also not sure how to proceed with the bb_pad_collate function.
So, I am a little bit lost at this point. Can anybody help me?
If so, the dataset does not have only one category: you have also negative samples, that is, images that do not contain signet ring cells. They are not annotated, but you can just extract samples of non-signet ring cells.
Hi @VDM,
Yes, I am using Digestpath as a start. Yes, it’s true that there are negative samples but, should I create random bounding boxes for the training? Or should I just take the negative samples and crop it like the ring cell images?
In the end, the metrics that should be used have to take into account the bounding boxes, right?
