search

ultralytics / yolov5

YOLOv5 🚀 in PyTorch > ONNX > CoreML > TFLite
https://docs.ultralytics.com
GNU Affero General Public License v3.0
48.64k stars 15.91k forks source link

EdgeTPU Conversion Output Change #8814

Closed vyang2968 closed 1 year ago

vyang2968 commented 1 year ago

Search before asking

Description

When exporting the yolov5s.pt to EdgeTPU with python export.py --weights yolov5s.pt --include edgetpu, it outputs yolov5s-int8_edgetpu.tflite. Inspecting with Netron, the model has 1 output with dimension [1,6300,85]. With Coral pretrained models, they have 4 outputs coming from a TFLite_Detection_PostProcess node. Just like the exported yolov5s model, the outputs are a StatefulPartitionedCall, but their output dimensions are also different. In the case of the TF2 SSD MobileNet V2 model, it has 4 ouputs with dimensions [1,20], [1], [1,20,4], and [1,20]. Such a format allows the model to easily be used with the edgetpu_detect_server command on the Dev Board and the pycoral library.

Use case

Users of Google's Dev Board and Dev Board Mini will enjoy a "plug and play" type model that can be easily used in the edgetpu_detect and edgetpu_detect_server command as the --model argument. It will be easily able to extract the inputs and outputs from it and detect objects in the video stream. In addition, users of Google's Pycoral library will have better access to its easy to use functions.

Additional

No response

Are you willing to submit a PR?

RyanHir commented 1 year ago

While this will not help with software designed around the expectation of TFLite_Detection_PostProcess being the output, this guide on StackOverflow (https://stackoverflow.com/questions/65824714/process-output-data-from-yolov5-tflite/65953473#65953473) shows how to parse the output tensor data. This should work with the pycoral library, except for the functions in pycoral.adapters.detect as they have the expectation that the model has TFLite_Detection_PostProcess.

glenn-jocher commented 1 year ago

@vyang2968 hi, thank you for your feature suggestion on how to improve YOLOv5 🚀!

The fastest and easiest way to incorporate your ideas into the official codebase is to submit a Pull Request (PR) implementing your idea, and if applicable providing before and after profiling/inference/training results to help us understand the improvement your feature provides. This allows us to directly see the changes in the code and to understand how they affect workflows and performance.

Please see our ✅ Contributing Guide to get started.

vyang2968 commented 1 year ago

Unfortunately, I am unable to submit a PR because as of now, I do not have the capabilities to do this.

github-actions[bot] commented 1 year ago

👋 Hello, this issue has been automatically marked as stale because it has not had recent activity. Please note it will be closed if no further activity occurs.

Access additional YOLOv5 🚀 resources:

Access additional Ultralytics ⚡ resources:

Feel free to inform us of any other issues you discover or feature requests that come to mind in the future. Pull Requests (PRs) are also always welcomed!

Thank you for your contributions to YOLOv5 🚀 and Vision AI ⭐!

glenn-jocher commented 8 months ago

@vyang2968 Understood, submitting a PR is voluntary. Meanwhile, were you able to find the resources provided helpful for parsing the output tensor data as a workaround for your use case? Let me know if you need any further assistance!