3 ms·
Often times a body transformation from one image to another results in a redefinition of the relative position and length of edges in a photo (as a person gets
by iluvmylife 9y ago
Often times a body transformation from one image to another results in a redefinition of the relative position and length of edges in a photo (as a person gets leaner).
Do you think that the L1 regularization used in pix2pix - which is intended to preserve low-frequency spacial patterns in the image pairs - will ultimately end up limiting how well cGANs can learn the body transformation image translation task?
- rayheberer 9y agoGreat question! My intuition tells me that L1 regularization could indeed be impacting the performance of the model, but for me personally, the only way to find out for certain would be to test this empirically.