yolo v8 on OAK-d

krishnashravan

I'm trying to get person detection using the yolov8 the nano model
i'm getting this error:
[1844301051D1420E00] [1.1] [22.191] [XLinkOut(4)] [error] Message has too much metadata (138492B) to serialize. Maximum is 51200B. Dropping message.

jakaskerl

Hi krishnashravan,
Could you please attach a MRE? Also are you using the yolo for spatial detections by any chance?

Thanks,
Jaka

ChrisCoutureDelValle

krishnashravan I typically get this error when I define the wrong number of classes

krishnashravan

parentDir = Path(file).parent

model = 'yolov8n.blob'

customModel = True

sizeX = 640

sizeY = 640

def init_pipeline():

pipeline = depthai.Pipeline()

cam_rgb = pipeline.createColorCamera()

cam_rgb.setResolution(depthai.ColorCameraProperties.SensorResolution.THE_4_K)

cam_rgb.setPreviewSize(640, 480)

cam_rgb.setInterleaved(False)

xout_rgb = pipeline.createXLinkOut()

xout_rgb.setStreamName("rgb")

cam_rgb.preview.link(xout_rgb.input)

manip1 = pipeline.createImageManip()

manip1.setMaxOutputFrameSize(1244160)

manip1.initialConfig.setResize(sizeX, sizeY)

cam_rgb.preview.link(manip1.inputImage)

# Next, we want a neural network that will produce the detections

detection_nn = pipeline.createMobileNetDetectionNetwork()

# Blob is the Neural Network file, compiled for MyriadX. It contains both the definition and weights of the model

# We're using a blobconverter tool to retreive the MobileNetSSD blob automatically from OpenVINO Model Zoo

# detection_nn.setBlobPath(blobconverter.from_zoo(name='mobilenet-ssd', shaves=6))

# Next, we filter out the detections that are below a confidence threshold. Confidence can be anywhere between <0..1>

detection_nn.setConfidenceThreshold(0.5)

# Next, we link the camera 'preview' output to the neural network detection input, so that it can produce detections

manip1.out.link(detection_nn.input)

if customModel is True:

	nnPath = str((parentDir / Path('../../data/' + model)).resolve().absolute())

	detection_nn.setBlobPath(nnPath)

	print("Custom Model" + nnPath + "Size: " + str(sizeX) + "x" + str(sizeY))

else:

	detection_nn.setBlobPath(blobconverter.from_zoo(name='person-detection-0106', shaves=6))

	print("Model from OpenVINO Zoo" + "Size: " + str(sizeX) + "x" + str(sizeY))

xout_nn = pipeline.createXLinkOut()

xout_nn.setStreamName("nn")

detection_nn.out.link(xout_nn.input)

return pipeline

this is just part of the code the full code is very large
basically i'm trying to get people count inside a zone but since the use case was for cctv type with people very far away so trying with yolo v8
and no i'm not using for the spatial detections.

erik

Hi krishnashravan ,
Did you follow our YoloV8 training/conversion/deployment notebook? https://github.com/luxonis/depthai-ml-training/blob/master/colab-notebooks/YoloV8_training.ipynb
Thanks, Erik

krishnashravan

Hello erik,

Since the predefined model with all the classes i ran the nano,small,medium on a video that i wanted to try so was used the official mode to export it to onnx and used the luxonis tool to get the blob file

Then changed the image size to run it

Might make a custom model later to get more accurate results if needed.

Or do i need to make a custom one in order to run it on the oak?

krishnashravan

I used labelmap of 20 or so classes

So that much be it

If by anychance u know the total number of classes in the yolov8 nd the list if possible

ChrisCoutureDelValle

krishnashravan I believe the base yolov8n trained on coco has 80 classes. I'm assuming you're not using a custom model, so in addition, unless you specified 640x480 in conversion I believe you have to set your preview size to 640x640 as well, to be consistent.
Classes for coco128
https://github.com/ultralytics/ultralytics/blob/main/ultralytics/datasets/coco128.yaml

krishnashravan

ChrisCoutureDelValle

I will try this and get back to u.

krishnashravan

ChrisCoutureDelValle
Hi tried both the approaches changing the preview window and classified all the classes in a labelmap all 80 classes still getting the same error.

krishnashravan

I'm trying a new approach that
https://pyimagesearch.com/2023/05/15/hand-gesture-recognition-with-yolov8-on-oak-d-in-near-real-time/
thats there to export the model
i converted the ONNX model into the openvino model format using the cli command
mo --input_model yolov8s.onnx
But when i'm trying to convert it into a blob file using http://blobconverter.luxonis.com/

i'm getting this issue any way to solve this?

erik

krishnashravan have you followed the https://github.com/luxonis/depthai-ml-training/blob/master/colab-notebooks/YoloV8_training.ipynb notebook?