Skip to content

[prototype] Minor speed and nit optimizations on Transform Classes #6837

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
datumbox merged 17 commits into from
Oct 27, 2022
Merged

[prototype] Minor speed and nit optimizations on Transform Classes #6837

Changes from 1 commit
Commits
Show all changes
17 commits
Select commit Hold shift + click to select a range
49f7e5a
Change random generator for ColorJitter.
datumbox Oct 25, 2022
5ea065c
Merge branch 'main' into prototype/class_speedup2
datumbox Oct 26, 2022
99b1685
Move `_convert_fill_arg` from runtime to constructor.
datumbox Oct 26, 2022
17d8184
Remove unnecessary TypeVars.
datumbox Oct 26, 2022
5e0be6e
Remove unnecessary casts
datumbox Oct 26, 2022
08ae56f
Update comments.
datumbox Oct 26, 2022
7b8be17
Minor code-quality changes on Geometical Transforms.
datumbox Oct 26, 2022
5f7e1ee
Fixing linter and other minor fixes.
datumbox Oct 26, 2022
b0b9b55
Change mitigation for mypy.`
datumbox Oct 26, 2022
ee31969
Fixing the tests.
datumbox Oct 26, 2022
88328f5
Fixing the tests.
datumbox Oct 26, 2022
e15f536
Fix linter
datumbox Oct 26, 2022
53f12bb
Restore dict copy.
datumbox Oct 26, 2022
843bcc9
Handling of defaultdicts
datumbox Oct 26, 2022
8e6af8d
restore int idiom
datumbox Oct 26, 2022
11af094
Update todo
datumbox Oct 26, 2022
4f90ce1
Merge branch 'main' into prototype/class_speedup2
datumbox Oct 27, 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
2 changes: 1 addition & 1 deletion torchvision/prototype/transforms/_color.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -51,7 +51,7 @@ def _check_input(

@staticmethod
def _generate_value(left: float, right: float) -> float:
return float(torch.distributions.Uniform(left, right).sample())
return torch.empty(1).uniform_(left, right).item()
Copy link
Contributor Author

@datumbox datumbox Oct 25, 2022

Choose a reason for hiding this comment

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

Switching to this random generator we get a performance boost on GPU. Moreover this option is JIT-scriptable (if on the future we decide to add support) and doesn't require to constantly initialize a distribution object as before:

[--------- ColorJitter cpu torch.float32 ---------]
                     |   old random  |   new random
1 threads: ----------------------------------------
      (3, 400, 400)  |       17      |       17    
6 threads: ----------------------------------------
      (3, 400, 400)  |       21      |       21    

Times are in milliseconds (ms).

[--------- ColorJitter cuda torch.float32 --------]
                     |   old random  |   new random
1 threads: ----------------------------------------
      (3, 400, 400)  |      1090     |      883    
6 threads: ----------------------------------------
      (3, 400, 400)  |      1090     |      882    

Times are in microseconds (us).


def _get_params(self, flat_inputs: List[Any]) -> Dict[str, Any]:
fn_idx = torch.randperm(4)
Expand Down