def add_image_transform(self, ctx):
image_width, image_height = ctx.image_transformer.output_size
if image_height < image_width:
sx, sy = image_width / image_height, 1
else:
sx, sy = 1, image_height / image_width
ctx.image_transformer.mm(mat3.stretch(sx, sy))
def _transform_image(self, image, inverse=False):
# Move principle point to origin
matrix = torch.DoubleTensor([[1, 0, -self.x0], [0, 1, -self.y0],
[0, 0, 1]])
# Apply transformations
matrix = self.matrix.mm(matrix)
# Restore principle point
ow, oh = self.dest_size.tolist()
matrix = self._mm(
mat3.translate(self.x0 * ow / self.orig_width,
self.y0 * oh / self.orig_height), matrix)
output_size = self.dest_size.round().int()
if inverse:
matrix = matrix.inverse()
output_size = torch.IntTensor([self.orig_width, self.orig_height])
# Scale up
matrix = self._mm(mat3.stretch(self.msaa), matrix)
# Apply affine image transformation
inv_matrix = matrix.inverse().contiguous()
image = image.transform(tuple(output_size * self.msaa), Image.AFFINE,
tuple(inv_matrix[0:2].view(6)), Image.BILINEAR)
# Scale down to output size
if self.msaa != 1:
image = image.resize(tuple(output_size), Image.ANTIALIAS)
return image
def add_image_transform(self, ctx):
ctx.image_transformer.mm(mat3.stretch(self.sx, self.sy))
def add_camera_transform(self, ctx):
ctx.camera_transformer.mm(mat3.stretch(self.sx, self.sy))
def add_image_transform(self, ctx):
old_dest_size = ctx.image_transformer.dest_size
new_dest_size = ctx.image_transformer.set_output_size(self.out_width, self.out_height)
scale = new_dest_size / old_dest_size
ctx.image_transformer.mm(mat3.stretch(scale[0], scale[1]))
def zoom(self, sx, sy):
self.mm(mat3.stretch(sx, sy))
self.sx *= sx
self.sy *= sy
