pytorc 2d的平移

tensorflow_addons 里面的tfa.image.transform(input_, transforms, interpolation)做平移很方便，但是pytorch版的平移版本太多了。正好做刚体配准的时候，感觉有一个版本的平移比较好用：

def torch_image_translate(input_, tx, ty, interpolation='nearest'):
    # got these parameters from solving the equations for pixel translations
    # on https://www.tensorflow.org/api_docs/python/tf/contrib/image/transform
    translation_matrix = torch.zeros([1, 3, 3], dtype=torch.float)
    translation_matrix[:, 0, 0] = 1.0
    translation_matrix[:, 1, 1] = 1.0
    translation_matrix[:, 0, 2] = -2*tx/(input_.size()[2]-1)
    translation_matrix[:, 1, 2] = -2*ty/(input_.size()[3]-1)
    translation_matrix[:, 2, 2] = 1.0
    grid = F.affine_grid(translation_matrix[:, 0:2, :], input_.size()).to(input_.device)
    wrp = F.grid_sample(input_, grid, mode=interpolation)
    return wrp

如果需要其他仿射变换，就自己修改仿射矩阵好了。