Oak-d Pro / YOLOv8

Hello @jakaskerl

The example code really helps me a lot ,Thank you,After modify my mode,It works,but the output frame is continuously change frame size from 320x320 to much bigger frame size and to 320x320,I don't know where I setting wrong,Can you help me check my code.

import cv2

import depthai as dai

import numpy as np

model_weights_path = "C:/catdog320.blob"
custom_model_blob = dai.OpenVINO.Blob(model_weights_path)

numClasses = 80

dim = next(iter(custom_model_blob.networkInputs.values())).dims
output_name, output_tenser = next(iter(custom_model_blob.networkOutputs.items()))

if "yolov6" in output_name:

    numClasses = output_tenser.dims[2] - 5

else:

    numClasses = output_tenser.dims[2] // 3 - 5
labelMap = [    "class_%s" % i

    for i in range(numClasses)]

pipeline = dai.Pipeline()

monoL = pipeline.create(dai.node.MonoCamera)

manip = pipeline.create(dai.node.ImageManip)

manipOut = pipeline.create(dai.node.XLinkOut)

detectionNetwork = pipeline.create(dai.node.YoloDetectionNetwork)

monoL.setNumFramesPool(24)

xoutNN = pipeline.create(dai.node.XLinkOut)

manipOut.setStreamName('flood-left')

xoutNN.setStreamName("nn")

# Script node for frame routing and IR dot/flood alternate

script = pipeline.create(dai.node.Script)

script.setProcessor(dai.ProcessorType.LEON_CSS)

script.setScript("""

    dotBright = 0.8

    floodBright = 0.1

    LOGGING = False  # Set True for latency/timings debugging
    node.warn(f'IR drivers detected: {str(Device.getIrDrivers())}')
    flagDot = False

    while True:        # Wait first for a frame event, received at MIPI start-of-frame        event = node.io['event'].get()

if LOGGING: tEvent = Clock.now()
  # Immediately reconfigure the IR driver.        # Note the logic is inverted, as it applies for next frame       

Device.setIrLaserDotProjectorIntensity(0 if flagDot else dotBright)       

Device.setIrFloodLightIntensity(floodBright if flagDot else 0)       

if LOGGING: tIrSet = Clock.now()
      # Wait for the actual frame (after MIPI capture and ISP proc is done) 

    frameL = node.io['frameL'].get()       

if LOGGING: tLeft = Clock.now()
    if LOGGING:

latIR      = (tIrSet - tEvent               ).total_seconds() * 1000   

    latEv      = (tEvent - event.getTimestamp() ).total_seconds() * 1000           

latProcL   = (tLeft  - event.getTimestamp() ).total_seconds() * 1000   

      node.warn(f'T[ms] latEv:{latEv:5.3f} latIR:{latIR:5.3f} latProcL:{latProcL:6.3f}')
      if not flagDot:            node.io['floodL'].send(frameL)
        flagDot = not flagDot""")

detectionNetwork.setBlob(custom_model_blob)

detectionNetwork.setConfidenceThreshold(0.4)
detectionNetwork.setNumClasses(numClasses)

detectionNetwork.setCoordinateSize(4)

detectionNetwork.setAnchors([])

detectionNetwork.setAnchorMasks({})

detectionNetwork.setIouThreshold(0.5)

detectionNetwork.input.setBlocking(False)
manip.initialConfig.setResize(320, 320)

manip.initialConfig.setFrameType(dai.ImgFrame.Type.BGR888p)

monoL.out.link(manip.inputImage)

manip.out.link(detectionNetwork.input)

manip.out.link(manipOut.input)

detectionNetwork.out.link(xoutNN.input)

monoL.frameEvent.link(script.inputs['event'])

monoL.out.link(script.inputs['frameL'])

script.outputs['floodL'].link(manipOut .input)

# Connect to device and start pipeline

with dai.Device(pipeline) as device:

    qRight = device.getOutputQueue("flood-left", maxSize=4, blocking=False)

    qDet = device.getOutputQueue("nn", maxSize=4, blocking=False)
    frame = None

    detections = []
    def frameNorm(frame, bbox): 

    normVals = np.full(len(bbox), frame.shape[0])   

    normVals[::2] = frame.shape[1]     

  return (np.clip(np.array(bbox), 0, 1) * normVals).astype(int)
    def drawText(frame, text, org, color=(255, 255, 255), thickness=1):

        cv2.putText(frame, text, org, cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 0), thickness + 3, cv2.LINE_AA) 

      cv2.putText(frame, text, org, cv2.FONT_HERSHEY_SIMPLEX, 0.5, color, thickness, cv2.LINE_AA        )

def drawRect(frame, topLeft, bottomRight, color=(255, 255, 255), thickness=1):       

cv2.rectangle(frame, topLeft, bottomRight, (0, 0, 0), thickness + 3)   

    cv2.rectangle(frame, topLeft, bottomRight, color, thickness)

def displayFrame(name, frame): 

      color = (128, 128, 128)     

  for detection in detections:     

      bbox = frameNorm(frame, (detection.xmin, detection.ymin, detection.xmax, detection.ymax) )            drawText( frame=frame, text=labelMap[detection.label],  org=(bbox[0] + 10, bbox[1] + 20), )            drawText(frame=frame,  text=f"{detection.confidence:.2%}", org=(bbox[0] + 10, bbox[1] + 35),)            drawRect(frame=frame,topLeft=(bbox[0], bbox[1]),bottomRight=(bbox[2], bbox[3]),color=color,)        cv2.putText(frame, '77777', (10, 30), cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 0, 0), 2)        cv2.imshow(name, frame)
while True:

        imageQueueData = qRight.tryGet()

        detectQueueData = qDet.tryGet()
        if imageQueueData is not None:

            frame = imageQueueData.getCvFrame()
        if detectQueueData is not None:

            detections = detectQueueData.detections
        if frame is not None:

            displayFrame("rgb", frame)
        if cv2.waitKey(1) == ord("q"):

            break

cv2.destroyAllWindows()

The text format may be wrong. Please understand.

Thanks,

LI