Hi!
I recently started working with the OAK-1 AF (USB). After having run the latency measurement code (https://docs.luxonis.com/projects/api/en/latest/samples/host_side/latency_measurement/#latency-measurement) successfully, I got the expected average latency between 30-40 ms with the FPS set to 10 and 60.
Next I combined this latency code with the code including yolo (examples/NeuralNetwork/detection_parser.py) and run it with 10 and 60 FPS, resulting in very different latencies:
Do you have an idea, why the latency is so much higher with higher FPS?
In my understanding the reading of the sensor should take the same time regardless of the FPS and especially Yolo's latency should stay the same since the image size does not change. Am I missing something completely?
Code:
from pathlib
import Path
import cv2
import depthai as dai
import numpy as np
import argparse
nnPath = str((Path(file).parent / Path('/Users/lasse/depthai-python/examples//models/mobilenet-ssd_openvino_2021.4_6shave.blob')).resolve().absolute())
labelMap = ["background", "aeroplane", "bicycle", "bird", "boat", "bottle", "bus", "car", "cat", "chair", "cow","diningtable", "dog", "horse", "motorbike", "person", "pottedplant", "sheep", "sofa", "train", "tvmonitor"]
Set up piplines and linking them
Create pipeline
pipeline = dai.Pipeline()
This might improve reducing the latency on some systems
pipeline.setXLinkChunkSize(0)
Define source and output
camRgb = pipeline.create(dai.node.ColorCamera)nn = pipeline.create(dai.node.NeuralNetwork)det = pipeline.create(dai.node.DetectionParser)
xoutRgb = pipeline.create(dai.node.XLinkOut)xoutRgb.setStreamName("rgb")nnOut = pipeline.create(dai.node.XLinkOut)nnOut.setStreamName("nn")
nn.passthrough.link(xoutRgb.input)
camRgb.setFps(60)camRgb.setPreviewSize(300, 300)camRgb.setInterleaved(False)camRgb.setResolution(dai.ColorCameraProperties.SensorResolution.THE_1080_P)camRgb.isp.link(xoutRgb.input)
Define a neural network that will make predictions based on the source frames
nn.setNumInferenceThreads(2)nn.input.setBlocking(False)
blob = dai.OpenVINO.Blob(nnPath)nn.setBlob(blob)det.setBlob(blob)det.setNNFamily(dai.DetectionNetworkType.MOBILENET)det.setConfidenceThreshold(0.5)
camRgb.preview.link(nn.input)nn.out.link(det.input)det.out.link(nnOut.input)
Connect to device and start pipeline
with dai.Device(pipeline) as device:
print(device.getUsbSpeed(), ' - FPS set to ', camRgb.getFps())
# Set queues maSize to 1 and blocking to false -> I am only interested in the most recent enrty
qRgb = device.getOutputQueue(name="rgb", maxSize=1, blocking=False)
qDet = device.getOutputQueue(name="nn", maxSize=1, blocking=False)
detections = []
N = 10
count = 0
diffs_img = np.array([])
diffs_nn = np.array([])
while count < N:
# get image from queue
imgFrame = qRgb.get()
latencyMs = (dai.Clock.now() - imgFrame.getTimestamp()).total_seconds() * 1000
diffs_img = np.append(diffs_img, latencyMs)
print('IMG: Latency: {:.2f} ms, Average latency: {:.2f} ms, Std: {:.2f}'.format(latencyMs, np.average(diffs_img), np.std(diffs_img)))
# get detection from queue
inDet = qDet.get()
latencyMs = (dai.Clock.now() - inDet.getTimestamp()).total_seconds() * 1000
diffs_nn = np.append(diffs_nn, latencyMs)
print('nn: Latency: {:.2f} ms, Average latency: {:.2f} ms, Std: {:.2f}'.format(latencyMs, np.average(diffs_nn), np.std(diffs_nn)))
# print out all detections
'''if inDet is not None:
detections = [detection for detection in inDet.detections]
for detection in detections:
print(labelMap[detection.label], detection.confidence * 100, ' %')'''
count += 1`
