Pytorch differentiable image transform operations

---

All functions are constructed on Pytorch basic tensor operations which supporting autograd backward functions. Transform functions are tested on CUDA available environment

Supporting operations

1. Affine transform
   • Rotation
   • Translate X
   • Translate Y
   • Sheer X
   • Sheer Y
2. Image enhancement
   • Contrast control
   • Brightness control

---

How to use

from transform_module import *

# create new sample image
ori_image = torch.randint((1,3,256,256)).cuda()

# create transform module
# doesn't need to cast transform module to gpu
# all operations conducted with pytorch functional
# takes pytorch tensor
transform_module = Rotation() 

# transform image
trans_image = transform_module(ori_image, param)