Skip to content

[prototype] Adjust solarize threshold on input type #6874

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 5 commits into from
Nov 1, 2022
Merged

[prototype] Adjust solarize threshold on input type #6874

Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
daf7893
Adjust solarize threshold on input type
datumbox Oct 31, 2022
9d75f23
Handle PIL images
datumbox Oct 31, 2022
ae0390e
minor refactor
datumbox Oct 31, 2022
d6f4fd7
Fix linter
datumbox Oct 31, 2022
6c5d74b
Merge branch 'main' into prototype/aa_solarize
datumbox Nov 1, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
19 changes: 7 additions & 12 deletions torchvision/prototype/transforms/_auto_augment.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -137,7 +137,8 @@ def _apply_image_or_video_transform(
elif transform_id == "Posterize":
return F.posterize(image, bits=int(magnitude))
elif transform_id == "Solarize":
return F.solarize(image, threshold=magnitude)
bound = 1.0 if isinstance(image, torch.Tensor) and image.is_floating_point() else 255.0
return F.solarize(image, threshold=bound * magnitude)
elif transform_id == "AutoContrast":
return F.autocontrast(image)
elif transform_id == "Equalize":
Expand Down Expand Up @@ -169,7 +170,7 @@ class AutoAugment(_AutoAugmentBase):
lambda num_bins, height, width: (8 - (torch.arange(num_bins) / ((num_bins - 1) / 4))).round().int(),
False,
),
"Solarize": (lambda num_bins, height, width: torch.linspace(255.0, 0.0, num_bins), False),
"Solarize": (lambda num_bins, height, width: torch.linspace(1.0, 0.0, num_bins), False),
"AutoContrast": (lambda num_bins, height, width: None, False),
"Equalize": (lambda num_bins, height, width: None, False),
"Invert": (lambda num_bins, height, width: None, False),
Expand Down Expand Up @@ -324,7 +325,7 @@ class RandAugment(_AutoAugmentBase):
lambda num_bins, height, width: (8 - (torch.arange(num_bins) / ((num_bins - 1) / 4))).round().int(),
False,
),
"Solarize": (lambda num_bins, height, width: torch.linspace(255.0, 0.0, num_bins), False),
"Solarize": (lambda num_bins, height, width: torch.linspace(1.0, 0.0, num_bins), False),
"AutoContrast": (lambda num_bins, height, width: None, False),
"Equalize": (lambda num_bins, height, width: None, False),
}
Expand Down Expand Up @@ -378,7 +379,7 @@ class TrivialAugmentWide(_AutoAugmentBase):
lambda num_bins, height, width: (8 - (torch.arange(num_bins) / ((num_bins - 1) / 6))).round().int(),
False,
),
"Solarize": (lambda num_bins, height, width: torch.linspace(255.0, 0.0, num_bins), False),
"Solarize": (lambda num_bins, height, width: torch.linspace(1.0, 0.0, num_bins), False),
"AutoContrast": (lambda num_bins, height, width: None, False),
"Equalize": (lambda num_bins, height, width: None, False),
}
Expand Down Expand Up @@ -423,7 +424,7 @@ class AugMix(_AutoAugmentBase):
lambda num_bins, height, width: (4 - (torch.arange(num_bins) / ((num_bins - 1) / 4))).round().int(),
False,
),
"Solarize": (lambda num_bins, height, width: torch.linspace(255.0, 0.0, num_bins), False),
"Solarize": (lambda num_bins, height, width: torch.linspace(1.0, 0.0, num_bins), False),
"AutoContrast": (lambda num_bins, height, width: None, False),
"Equalize": (lambda num_bins, height, width: None, False),
}
Expand Down Expand Up @@ -505,13 +506,7 @@ def forward(self, *inputs: Any) -> Any:
aug = self._apply_image_or_video_transform(
aug, transform_id, magnitude, interpolation=self.interpolation, fill=self.fill
)
mix.add_(
# The multiplication below could become in-place provided `aug is not batch and aug.is_floating_point()`
# Currently we can't do this because `aug` has to be `unint8` to support ops like `equalize`.
# TODO: change this once all ops in `F` support floats. https://github.com/pytorch/vision/issues/6840
combined_weights[:, i].reshape(batch_dims)
* aug
)
mix.add_(combined_weights[:, i].reshape(batch_dims) * aug)
mix = mix.reshape(orig_dims).to(dtype=image_or_video.dtype)

if isinstance(orig_image_or_video, (features.Image, features.Video)):
Expand Down