Closed saandeepa93 closed 4 years ago
Pytorch grid_warp method assumes that your coordinated is in [-1,1] range. So you can either use this range from the begining, like I did or use [0,W]x[0,H] range and then normalise when you pass to grid_warp. These are equivalent.
Got it! Thank you for your quick response.
Hello, What is the intuition behind squishing the grid coordinates between [-1, 1] before calculating the expectation and covariance of the heatmaps? In the paper it is mentioned that the grid coordinates have values within the H*W coordinates but in the code those values are reduced to [-1, 1]. Referring to the make_coordinate_grid method under /modules/util.py. Please let me know if I am missing something.
Thanks in advance!