replace tensor division with scalar division and tensor multiplication (#6903)

* replace tensor division with scalar division and tensor multiplication * fix consistency test tolerances

Author: pmeier

test/test_prototype_transforms_consistency.py

@@ -163,6 +163,8 @@ def __init__(
  ArgsKwargs(torch.uint8),
  ],
  supports_pil=False,
+ # Use default tolerances of `torch.testing.assert_close`
+ closeness_kwargs=dict(rtol=None, atol=None),
  ),
  ConsistencyConfig(
  prototype_transforms.ToPILImage,

torchvision/prototype/transforms/functional/_color.py

@@ -180,7 +180,7 @@ def _rgb_to_hsv(image: torch.Tensor) -> torch.Tensor:
  hb = (4.0 + gc).sub_(rc).mul_(mask_maxc_neq_g & mask_maxc_neq_r)
 
  h = hr.add_(hg).add_(hb)
- h = h.div_(6.0).add_(1.0).fmod_(1.0)
+ h = h.mul_(1.0 / 6.0).add_(1.0).fmod_(1.0)
  return torch.stack((h, s, maxc), dim=-3)
 
 
@@ -287,7 +287,7 @@ def adjust_gamma(inpt: features.InputTypeJIT, gamma: float, gain: float = 1) ->
 def posterize_image_tensor(image: torch.Tensor, bits: int) -> torch.Tensor:
  if image.is_floating_point():
  levels = 1 << bits
- return image.mul(levels).floor_().clamp_(0, levels - 1).div_(levels)
+ return image.mul(levels).floor_().clamp_(0, levels - 1).mul_(1.0 / levels)
  else:
  num_value_bits = _num_value_bits(image.dtype)
  if bits >= num_value_bits:

torchvision/prototype/transforms/functional/_meta.py

@@ -367,7 +367,7 @@ def convert_dtype_image_tensor(image: torch.Tensor, dtype: torch.dtype = torch.f
  else:
  # int to float
  if float_output:
- return image.to(dtype).div_(_FT._max_value(image.dtype))
+ return image.to(dtype).mul_(1.0 / _FT._max_value(image.dtype))
 
  # int to int
  num_value_bits_input = _num_value_bits(image.dtype)