Skip to content

Rem runner loops; use OpenCV #2

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.

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

Open
wants to merge 6 commits into
base: main
Choose a base branch
from

Conversation

seth-planet
Copy link
Contributor

@seth-planet seth-planet commented Jun 17, 2024

  • Remove TPU Runner loop threads. Instead just have TPUs in queues when ready for usage. Allows for better optimizations and fewer thread context switches. Runs faster with smaller models, up to 700 FPS with several TPUs. Also enables more types of optimizations. Now we track which TPU runs a particular segment faster and swap segments between TPUs to reduce bottlenecks.

  • Use OpenCV when installed. WILL NEED ADDITIONAL WORK TO VERIFY GOOD FALLBACK FUNCTIONALITY TO PIL. OpenCV is a large package, but contains good AVX optimizations for much faster image resizing.

  • Latest version of segmented files, now including YOLOv9. Includes new testing of multiple input tensor sizes to get around TPU internal memory. This will need work to ensure the expected files are in the expected places with the expected names.

New versions of archived TPU files available here:
https://drive.google.com/file/d/136o-0C8Z8G5ZUqq1Z_9wQV2KsjVSOZk6/view?usp=drive_link

@seth-planet
Copy link
Contributor Author

Coral tflite models have been uploaded to Hugging Face here:
https://huggingface.co/mailseth/coral/tree/main

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.

2 participants