Hi
maybe this question is not so related to pytroch but i believe this community can help me
i have a dataset where my ground truth images are so black and gray when i train the network on this dataset i am getting 100% accuracy because it read it as a black image
What question do you have regarding the image?
Do you encounter any errors in your code or would you just like to discuss the use case?
My question is if there is a way to make gt images more visible , with gray and black color it is hard to do the segmentation , the network does not seem to know where is the segmentted part and even when I display the image after the loader I am getting a completely balck image
How are you visualizing the target image and what kind of pixel values does it contain?
If you are dealing with a segmentation use case, each pixel should contain the corresponding class index. In your case if looks like the target has only two classes, thus it should only contain pixel values of 0 and 1.
matplotlib.pyplot usually scaled the colormap, so that you should get a black and white image.
Yes my case is segmentation , but will this quality of image effect the segmentation task or not , because I do not seem to make the model works with this dataset