Motivation

The default data type generated by np.ones is float64, which causes the data type of heatmap_weights to be float64 as well. Finally, it affects the efficiency of calculating KeypointMSELoss.

Modification

Modify the dtype of np.ones to speed up subsequent calculations.

Use cases (Optional)

We have verified on related configurations, such as configs/body_2d_keypoint/dekr/coco/dekr_hrnet-w32_8xb10-140e_coco-512x512.py, there is no accuracy problem after modification, and the performance is improved by about 1.5%

Environment information: Tesla V100, torch 2.0.0

Checklist

Before PR:

• [√] I have read and followed the workflow indicated in the CONTRIBUTING.md to create this PR.
• [√] Pre-commit or linting tools indicated in CONTRIBUTING.md are used to fix the potential lint issues.
• [√] Bug fixes are covered by unit tests, the case that causes the bug should be added in the unit tests.
• [√] New functionalities are covered by complete unit tests. If not, please add more unit tests to ensure correctness.
• [√] The documentation has been modified accordingly, including docstring or example tutorials.

After PR:

• [√] CLA has been signed and all committers have signed the CLA in this PR.