Skip to content

Allow register_kernel() to take dispatcher name as input #7796

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
NicolasHug merged 3 commits into from
Aug 3, 2023
Merged

Allow register_kernel() to take dispatcher name as input #7796

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
43a71e7
Allow register_kernel to take name as input
NicolasHug Aug 2, 2023
99ea401
Better _name_to_dispatcher
NicolasHug Aug 3, 2023
12fb8e1
Merge branch 'main' of github.com:pytorch/vision into register_str
NicolasHug Aug 3, 2023
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
33 changes: 33 additions & 0 deletions test/test_transforms_v2_refactored.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2181,3 +2181,36 @@ def test_unsupported_types(self, dispatcher, make_input):

with pytest.raises(TypeError, match=re.escape(str(type(input)))):
dispatcher(input)


class TestRegisterKernel:
@pytest.mark.parametrize("dispatcher", (F.resize, "resize"))
def test_register_kernel(self, dispatcher):
class CustomDatapoint(datapoints.Datapoint):
pass

kernel_was_called = False

@F.register_kernel(dispatcher, CustomDatapoint)
def new_resize(dp, *args, **kwargs):
nonlocal kernel_was_called
kernel_was_called = True
return dp

t = transforms.Resize(size=(224, 224), antialias=True)

my_dp = CustomDatapoint(torch.rand(3, 10, 10))
out = t(my_dp)
assert out is my_dp
assert kernel_was_called

# Sanity check to make sure we didn't override the kernel of other types
t(torch.rand(3, 10, 10)).shape == (3, 224, 224)
t(datapoints.Image(torch.rand(3, 10, 10))).shape == (3, 224, 224)

def test_bad_disaptcher_name(self):
class CustomDatapoint(datapoints.Datapoint):
pass

with pytest.raises(ValueError, match="Could not find dispatcher with name"):
F.register_kernel("bad_name", CustomDatapoint)
15 changes: 15 additions & 0 deletions torchvision/transforms/v2/functional/_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -37,7 +37,22 @@ def decorator(kernel):
return decorator


def _name_to_dispatcher(name):
import torchvision.transforms.v2.functional # noqa

try:
return next(
Copy link
Member Author

@NicolasHug NicolasHug Aug 2, 2023

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Not the most efficient, we could store the mapping somewhere. I don't think we care anyway since this is just executed during registration, not when calling the dispatcher.

Copy link
Contributor

@pmeier pmeier Aug 2, 2023

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Not sure I understand the iteration here. Can't we just do

try:
    return getattr(torchvision.transforms.v2.functional, name)
except AttributeError:
    raise ValueError(...) from None

Copy link
Member Author

@NicolasHug NicolasHug Aug 3, 2023

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

yes much better!

obj
for obj in torchvision.transforms.v2.functional.__dict__.values()
if getattr(obj, "__name__", "") == name
)
except StopIteration:
raise ValueError(f"Could not find dispatcher with name '{name}'.")


def register_kernel(dispatcher, datapoint_cls):
if isinstance(dispatcher, str):
dispatcher = _name_to_dispatcher(name=dispatcher)
return _register_kernel_internal(dispatcher, datapoint_cls, datapoint_wrapper=False)


Expand Down