Closed jianjun0407 closed 10 months ago
Point label 1 for positive points. Point label 0 for negative points. Point label -1 for ignored points.
Ignored points are used to fill the point prompt tensor into the same shape.
Thank you for your reply. I also have a question is, when the current algorithm is testing the image, I see that the pre-processing stage does not carry out a unified spacing operation, is there no need to deal with it? In other words, when you train the model, do you unify the spacing to a fixed value for data of different spacing?
No need to deal with it. We did not conduct the unified spacing operation during training phase and inference phase. This is due to the data scalability considerations.
Thank you for sharing. I have a question about the prompt, when the prompt is a point, there's no doubt that the label of the positive sample point is set to 1, so if it's a negative sample point, is its label set to -1, or is it set to 0?