cuda-gridsample-grad2
cuda-gridsample-grad2 copied to clipboard
Cuda implementation for gridsample with second derivative support
Cuda-based grid sample implimentation with second order derivative support.
For the forward and backward pass it uses the pytorch version of grid sample. Only the second backward is implimented here.
Tested only with Pytorch 10.1 and 10.2
Reflection padding is not supported
Usage:
import cuda_gridsample as cu
cu.grid_sample_2d(image, optical, padding_mode='border', align_corners=True)
cu.grid_sample_3d(volume, optical, padding_mode='border', align_corners=True)