[BUG] TypeError: send(): incompatible function arguments

[TomasMendozaHN]

trafficstars

(UPDATE): Please look at the second entry below. This error was fixed by wrapping my message with a default NN message type.

Hello :) First of all, I apologize if this is repeated. I also understand this is probably not a bug, but I am not sure.

So, I trained my own .blob NN, which takes an Array as input. I created the following script to send an array to the device:

# Create Pipeline
p = dai.Pipeline()
p.setOpenVINOVersion(dai.OpenVINO.VERSION_2021_4)

# Load my own NN
nn = p.createNeuralNetwork()
nn.setBlobPath(str(Path("resources/nn/carREID/Car.blob").resolve().absolute()))
nn.setNumInferenceThreads(2)
nn.input.setBlocking(True)

# Define XLink
xinArray = p.createXLinkIn()
nnOut = p.createXLinkOut()
xinArray.setStreamName("inArray")
nnOut.setStreamName("nn")

# Link everything
xinArray.out.link(nn.input)
nn.out.link(nnOut.input)

# Initialize everything
with dai.Device(p) as device:

    # Input queue to send arrays to the device.
    qIn = device.getInputQueue(name="inArray")
    
    # Output queue will be used to get nn data from the arrays.
    qDet = device.getOutputQueue(name="nn", maxSize=4, blocking=False)
    
    # Start!
    while True:

        # Create random array 
        x = torch.randn(1,3,64,128).numpy()
        
        # Send array to device and obtain result
        qIn.send(x)
        inDet = qDet.tryGet()

However, If I run such script, I get the following error:

TypeError: send(): incompatible function arguments. The following argument types are supported:

      1. (self: depthai.DataInputQueue, msg: depthai.ADatatype) -> None
      2. (self: depthai.DataInputQueue, rawMsg: depthai.RawBuffer) -> None

I understand that I need to transform my array, but I don't understand how. Thank you for your help :) :)

[TomasMendozaHN]

Hello again,

I have used the NN message module to wrap my message. However, I get a new error I don't understand. Here you can see my updated code:

from pathlib import Path
import numpy as np
import cv2
import depthai as dai
import time
import torch

p = dai.Pipeline()
p.setOpenVINOVersion(dai.OpenVINO.VERSION_2021_4)

nn = p.createNeuralNetwork()
nn.setBlobPath(str(Path("resources/nn/carREID/Car.blob").resolve().absolute()))
nn.setNumInferenceThreads(2)
nn.input.setBlocking(True)

xinArray = p.createXLinkIn()
nnOut = p.createXLinkOut()

xinArray.setStreamName("inArray")
nnOut.setStreamName("nn")

# Linking
xinArray.out.link(nn.input)
nn.out.link(nnOut.input)

with dai.Device(p) as device:

    # Input queue to send arrays to the device.
    qIn = device.getInputQueue(name="inArray")
    
    # Output queue will be used to get nn data from the arrays.
    qDet = device.getOutputQueue(name="nn", maxSize=84, blocking=False)
    
    counter = 0
    
    while True:

        x = torch.randn(1,3,64,128)

        # Now I convert my message here
        msg = dai.NNData()
        msg.setData(x)

       # send to NN to obtain result
        qIn.send(msg)
        inDet = qDet.tryGet()
        

    if inDet is not None:
        print('inDet: ', inDet)`

However, when running the above I get the following error:

 [NeuralNetwork(0)] [error] Input tensor 'input' (0) exceeds available data range. Data size (0B), tensor offset (0), size (24576B) - skipping inference

Does anyone know how to fix it? I think it has something to do with the input data flow? But I can't find anything online. Please help

[VanDavv]

Hi @TomasMendozaHN, could you change your x variable to be a planar list (so 1 dimension, plain python list object) and try again? I don't think NNData.setData would accept the torch.Tensor object that is returned from randn