ibaiGorordo / ONNX-YOLOv8-Object-Detection

Python scripts performing object detection using the YOLOv8 model in ONNX.
MIT License
378 stars 95 forks source link

Please add DeepSparse support #15

Open kopyl opened 1 year ago

kopyl commented 1 year ago

Your inference turned out to be much more accurate than Ultralytic's inference.

I found this interesting thing called DeepSparse. It works faster than ONNX but unfortunately its inference is less accurate than yours + uses heavyweight Torch.

https://github.com/neuralmagic/deepsparse

Could you please help me to use DeepSparse instead of ONNX?

kopyl commented 1 year ago

@ibaiGorordo i made it. Hard-coded some dimensions tho: https://pastebin.com/y2J32M90

Could you please help me make it without hard-coded values?

ibaiGorordo commented 1 year ago

My recommendation is to copy this repository, rename it to something like Deepsparse-YOLOv8-Object-Detection and change the yolov8 class with your changes. For easier findability.

kopyl commented 1 year ago

@ibaiGorordo that's what I did. But i suggest you to add the support :)

SeeknnDestroy commented 7 months ago

@kopyl how did you come up with the result of this repo gives higher accuracy than ultralytics yolo? can you elaborate?