How to use Yolov5 tflite?

[ngotra2710]

❔Question

How to set the input data before feed it into the tflite interpreter to get the correct output?

Additional context

Currently, I am trying to convert pytorch model file to tflite file. I has already successfully converted.

The tflite file and the input output details are in the zip file.
best_v2.zip

This is my code when I try to feed the input into the model
frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
if input_details[0]['dtype'] == np.float32:
floating_model = True
height = input_details[0]['shape'][2]
width = input_details[0]['shape'][3]
frame_resized = cv2.resize(frame_rgb, (width, height))
input_data = frame_resized if floating_model:
# input_data = (np.float32(input_data) - 127.5) / 127.5
input_data = (np.float32(input_data) / 255.0)
input_data = np.reshape(np.expand_dims(input_data, 0), (input_details[0]['shape']))
interpreter.set_tensor(input_details[0]['index'], input_data)
interpreter.invoke()
output_data = interpreter.get_tensor(output_details[0]['index'])
reshape_output_data = np.reshape(output_data, (output_data.shape[0],
output_data.shape[1]*output_data.shape[2]*output_data.shape[3], output_data.shape[4]))

When I try to "0 - 255 to 0.0 - 1.0" I got this output result:
[[[ 2.1864 0.77372 1.0798 0.87739 -11.606 3.5788]
[ 1.0578 0.83459 1.5088 0.64846 -10.49 3.8019]
[ 0.39111 0.40996 2.2016 0.33661 -12.224 3.9996]
...
[ 0.11459 -1.668 0.43695 -0.27363 -12.781 4.5359]
[ -0.88718 -1.2919 -0.11859 -0.20273 -14.401 4.745]
[ -0.87244 -0.45423 -0.1573 -0.03495 -16.105 4.8643]]]

What I want is: (same with the result in detect.py)
[tensor([[ 65.62970, 234.14124, 154.66338, 319.21924, 0.95002, 0.00000],
[425.51935, 203.04468, 528.26880, 267.44305, 0.93659, 0.00000],
[326.86328, 99.11230, 455.14697, 209.28619, 0.91681, 0.00000]])]

Can someone know about how to use the tflite, please explain it for me?
I am thinking that maybe my tflite was not converted correctly but I can not tell. Because I just follow step-by-step with the command line. (pt -> onnx -> pb ->tflite)

[github-actions]

Hello @ngotra2710, thank you for your interest in our work! Please visit our Custom Training Tutorial to get started, and see our Jupyter Notebook , Docker Image, and Google Cloud Quickstart Guide for example environments.

If this is a bug report, please provide screenshots and minimum viable code to reproduce your issue, otherwise we can not help you.

If this is a custom model or data training question, please note Ultralytics does not provide free personal support. As a leader in vision ML and AI, we do offer professional consulting, from simple expert advice up to delivery of fully customized, end-to-end production solutions for our clients, such as:

• Cloud-based AI systems operating on hundreds of HD video streams in realtime.
• Edge AI integrated into custom iOS and Android apps for realtime 30 FPS video inference.
• Custom data training, hyperparameter evolution, and model exportation to any destination.

For more information please visit https://www.ultralytics.com.

[glenn-jocher]

@ngotra2710 #959 may be of use to you.

[github-actions]

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.

[tino926]

@ngotra2710 when you convert to .onnx, You lost the operations to do sigmoid and rectify values according to anchors. Try implement those steps following yolo.py->class Detect->forward().

On the other hand, could you share how you convert .onnx to .pb? I tried using https://github.com/onnx/onnx-tensorflow to do this, but I got this error:
error: 'tf.AddV2' op is neither a custom op nor a flex op