Slow inference on static images

fransay commented 11 months ago

I am looking at a way, we could use your implementation on live object detection. But looks like inference on static images alone are quite slow. Will be learning more about your design and algorithm and see we can improve it.

Great job the way :+1:

JobiJoba commented 11 months ago

I would be interested by doing that too but I'm a bit afraid of the slowness of the static image right now.

ferraridamiano commented 11 months ago

Yeah, it is true, the current implementation takes really too much time. And I'm not even sure which could be the issue. It could be:

YOLO (e.g. the model is not well optimized to run on tflite)
The tflite flutter library (maybe not yet very optimized)
The app (maybe I did not enabled some option that unlocks more performance)

While the NMS algorithm could be easily speeded up, and maybe I will do it in the near future. But at the moment it takes only a fraction of time of the inference time

Feel free to open PRs that fix this problem

ferraridamiano commented 10 months ago

Seems like a flutter_tflite issue: https://github.com/tensorflow/flutter-tflite/issues/112

ferraridamiano / yolo_flutter

Slow inference on static images #2