Hi everyone, i am curious about scaling in keypoint normalization. Is there any specific reason for choosing the scaling value in this way? I mean why to move it to [-0.7; 0.7] range and not [-1.0; 1.0] for example?
https://github.com/magicleap/SuperGluePretrainedNetwork/blob/master/models/superglue.py#L71