[Support]: Yolov5 as TFLITE model works?

[ozett]

Describe the problem you are having

i am wondering if anybody had tried to convert yolov5 to tflite and got it running with frigate.
any experience with this and is it worth a try?

Version

DEBUG 0.10.0-B912851

here is how this could be done:

ultralytics/yolov5#4586

Install method

Docker CLI

Coral version

M.2

Network connection

Wired

Camera make and model

all hikvision

Any other information that may be helpful

[ozett]

i did it ...

python3 export.py --weights yolov5s.pt --include tflite

[ozett]

i missed a step?

[ozett]

convertion was fast as hell...

[ozett]

same error...

[ozett]

another try in colab..

[ozett]

something is missing...

[ozett]

??
google-coral/edgetpu#419 (comment)

[mr6880]

Same issue here when I tried it a few days ago

Process detector:coral_pci:
Traceback (most recent call last):
File "/usr/lib/python3.8/multiprocessing/process.py", line 315, in _bootstrap
self.run()
File "/usr/lib/python3.8/multiprocessing/process.py", line 108, in run
self._target(*self._args, **self._kwargs)
File "/opt/frigate/frigate/edgetpu.py", line 180, in run_detector
detections = object_detector.detect_raw(input_frame)
File "/opt/frigate/frigate/edgetpu.py", line 110, in detect_raw
self.interpreter.tensor(self.tensor_output_details[3]["index"])()[0]
IndexError: list index out of range
Process detector:coral_pci3:
Traceback (most recent call last):
File "/usr/lib/python3.8/multiprocessing/process.py", line 315, in _bootstrap
self.run()
File "/usr/lib/python3.8/multiprocessing/process.py", line 108, in run
self._target(*self._args, **self._kwargs)
File "/opt/frigate/frigate/edgetpu.py", line 180, in run_detector
detections = object_detector.detect_raw(input_frame)
File "/opt/frigate/frigate/edgetpu.py", line 110, in detect_raw
self.interpreter.tensor(self.tensor_output_details[3]["index"])()[0]
IndexError: list index out of range
Process detector:coral_pci2:
Traceback (most recent call last):
File "/usr/lib/python3.8/multiprocessing/process.py", line 315, in _bootstrap
self.run()
File "/usr/lib/python3.8/multiprocessing/process.py", line 108, in run
self._target(*self._args, **self._kwargs)
File "/opt/frigate/frigate/edgetpu.py", line 180, in run_detector
detections = object_detector.detect_raw(input_frame)
File "/opt/frigate/frigate/edgetpu.py", line 110, in detect_raw
self.interpreter.tensor(self.tensor_output_details[3]["index"])()[0]
IndexError: list index out of range

[rogerquake]

Haven't tried yet, but curious how the results were after importing into Frigate? If it worked good, it would be amazing to get it uploaded so we didn't have to compile it!

[ozett]

happens at the point where the model is loaded, i guess
looks like the converted model has different parameters than the usual edgetpu-models from google...

frigate/frigate/edgetpu.py

Lines 102 to 111 in a2d1bd2

	def detect_raw(self, tensor_input):
	self.interpreter.set_tensor(self.tensor_input_details[0]["index"], tensor_input)
	self.interpreter.invoke()
	boxes = self.interpreter.tensor(self.tensor_output_details[0]["index"])()[0]
	class_ids = self.interpreter.tensor(self.tensor_output_details[1]["index"])()[0]
	scores = self.interpreter.tensor(self.tensor_output_details[2]["index"])()[0]
	count = int(
	self.interpreter.tensor(self.tensor_output_details[3]["index"])()[0]
	)

edit: seems a well known problem: google-coral/edgetpu#272 (comment)

[ozett]

google-coral/edgetpu#272

[ozett]

😱

Thanks. My colleague run YOLOv5 v4 320x320 on EdgeTPU for 24 hours yesterday without errors.
google-coral/edgetpu#272 (comment)

edit: but as June 2021 still unsolved issue.. -> google-coral/edgetpu#272 (comment)

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.