Skip to content

Conversation

glenn-jocher
Copy link
Member

@glenn-jocher glenn-jocher commented Nov 14, 2022

@Laughing-q @AyushExel copy-paste zero-value fix here, verified working.

train_batch0

Signed-off-by: Glenn Jocher [email protected]

πŸ› οΈ PR Summary

Made with ❀️ by Ultralytics Actions

🌟 Summary

Improved Copy-Paste augmentation logic for better training efficiency in YOLOv5.

πŸ“Š Key Changes

  • Changed the color filled in contours from white (255, 255, 255) to (1, 1, 1) during the Copy-Paste augmentation process.
  • Simplified the combination of the original and augmented images by removing the bitwise_and operation and redundant pixel replacement logic.
  • Retained only the essential image flipping operation and updated how the augmented image pixels are combined with the original.

🎯 Purpose & Impact

  • 🎯 The adjustments streamline the Copy-Paste augmentation, potentially improving the training speed and efficiency.
  • πŸš€ Users might notice faster training times due to the more straightforward augmentation process.
  • πŸ“‰ The change could reduce computational overhead, indirectly benefiting users with less powerful hardware.

Signed-off-by: Glenn Jocher <[email protected]>
@glenn-jocher glenn-jocher self-assigned this Nov 14, 2022
Signed-off-by: Glenn Jocher <[email protected]>
@glenn-jocher glenn-jocher merged commit abbfd69 into master Nov 14, 2022
@glenn-jocher glenn-jocher deleted the glenn-jocher-patch-3 branch November 14, 2022 17:54
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant