By: liuyanice098 on Sept. 14, 2022, 9:26 a.m.
For the GT problem, the GT itself is a 64 bit TIFF gray-scale image. When training, I need to save the GT in an 8-bit png format. My prediction image is also an 8-bit png image with 3 channels . Then , I change the 3 channels png image to 1 channel gray-scale image. Finally, I save a 32-bit tiff gray-scale image. Is that OK？
By: liuyanice098 on Sept. 14, 2022, 9:30 a.m.
Does GT need to use 8-bit color images ?
By: junma on Sept. 14, 2022, 12:53 p.m.
The GT of training images is already available. They are 8-bit or 16-bit one-channel image depending the number of cells (e.g., 8-bit image can only save 255 cells. If the image contains more than 255 cells, we need to use 16-bit).
I guess you are talking about segmentation task. It should be a one-channel image (the same as the groud truth format). Please look at the baseline for more details.
