some questions on the control points #2
Description
Thanks for this repo!
I have some questions on the prediction of control points.
(1). Why the image_size inputted to the decoder is (112 * 112)?
(2). What's the use of ldmark_mean?
(3). How to visualize the control points? (I have tried to visualize the control points by outputting Decoder/WarpController/ldmark_pred and Decoder/WarpController/ldmark_diff, but some of them are out of the image. What's more, when using different warping scale, the output image changes, but ldmark_diff remains constant...)