Computer Vision News 42 CVPR Poster Presentation that directly manipulating the points within the data was crucial to achieving the desired outcomes, and it was necessary to establish a correspondence between the generated data and the reference data to accomplish this. When observing an image composed of points, specific characteristics, such as holes, become apparent. These holes can be matched based on their size, the distribution of different cell types, or the contextual information surrounding them. Shahira employed a matching technique based on persistence , which correlated with a cell’s size and the cross K-function around the holes. This matching approach established the desired spatial context. “ Now that we have a correspondence, we want the matched or paired locations to have a similar distribution of cells around them, ” Shahira points out. “ We want a loss function that allows us to move points around to affect the points in general. We createdmulti-scale density maps from the point maps for each class and tried to minimize the distance at the matched locations. This way, you will bring points together at the paired locations or move them apart to have the same density values. ”
RkJQdWJsaXNoZXIy NTc3NzU=