About the pseudo-masks produced by MaskCut #73
Description
For my dataset, I used demo.py to visualize the generated pseudo labels, and found that the quality of the generated labels was very poor. How can I improve the quality of the generated pseudo labels?
In addition, the hyperparameter N is the number of pseudo labels generated from an image, but the number of instances in each image is not exactly the same. How to deal with this? For example, the instances number of a image are 2, and another image are 3, but we use the N=3 to generate pseudo labels both.
Looking forward to hearing from you.