mvncStatus.TIMEOUT

[idata]

I get" raise Exception(Status(status)) Exception: mvncStatus.TIMEOUT" when I run graph.GetResult(), does anyone know why?

from mvnc import mvncapi as mvnc import numpy as np import cv2 import os import sys import tensorflow as tf IMAGE_PATH =os.path.dirname(os.path.realpath(__file__))+'/disparity.png' GRAPH_PATH=os.path.dirname(os.path.realpath(__file__))+'/model/CCNN.graph' PAD=4 def do_initialize() -> (mvnc.Device, mvnc.Graph): # *************************************************************** # Get a list of ALL the sticks that are plugged in # *************************************************************** devices = mvnc.EnumerateDevices() if len(devices) == 0: print('Error - No devices found') return (None, None) device.OpenDevice() # Load graph file try : with open(GRAPH_PATH, mode='rb') as f: in_memory_graph = f.read() except : print ("Error reading graph file: " + graph_filename) graph = device.AllocateGraph(in_memory_graph) return device, graph def forward(graph: mvnc.Graph, image_filename: str): print("[*] Forwarding....") # Load image from disk and preprocess it to prepare it for the network disparity = np.expand_dims(np.expand_dims(cv2.imread(image_filename, cv2.IMREAD_GRAYSCALE), 0), -1) padded_disparity = np.pad(disparity, ((0, 0), (PAD, PAD), (PAD, PAD), (0, 0)), "reflect") # Start the inference by sending to the device/graph graph.LoadTensor(padded_disparity, "user object for this inference") output, userobj = graph.GetResult() result = (output*65535.).astype(np.uint16) cv2.imwrite('test.png', result[0])

[idata]

@tommy Which model are you using with this code? I'd like to reproduce the issue with your code if you can provide a link to the model.

Is your device still visible in the system after the timeout issue using the lsusb command? Looks like there may be a main() that is missing, that would be helpful to have also. Thanks.

[idata]

Here you can find the model and the main too: https://github.com/tommyliverani/CCNN

Yes the device is still visible after the timeout and i have the same issue using mvNCCheck and mvNCProfile

[Error 25] Myriad Error: "mvncStatus.TIMEOUT".

[idata]

@Tome_at_Intel In your opinion, is there any problem in the main?

[idata]

@tommy I'm trying to recompile your graph file so that I can reproduce your issue, but I'm not able to. The command I'm using is mvNCCompile CCNN.meta -s 12 -on CCNN/prediction/conv/BiasAdd and I'm running into an out of index error. Can you share how you compiled your graph file?

[idata]

@Tome_at_Intel mvNCCompile CCNN.meta -s 12 -in=input -on=CCNN/prediction/conv/BiasAdd

[idata]

@Tome_at_Intel

I discovered that the problems are in 3x3 convolutions. Are that layers supported?

[idata]

@tommy 3x3 convolutions are supported (See release notes at https://developer.movidius.com/docs).

The command you used isn't working for me. Which version of the NCSDK are you using?

[idata]

@Tome_at_Intel ncsdk-1.12.00.01

[idata]

@Tome_at_Intel I've discovered with mvncProfile that the layers that raise the time out have a higher inference time that the other networks i've worked on. Could this be the problem?

[idata]

@tommy If the output of the model is too large, it could cause the timeout issue in addition to nans in some intermediate layers. You can try adding an epsilon value to scale the values down in the batchnorm layer. For example:

layer { name: "BatchNorm1" type: "BatchNorm" bottom: "Convolution1" top: "Convolution1" batch_norm_param { eps: 1e-2 } }

You can also use mvNCCheck with the -on option to check the outputs of each layer of your network as well. More information on mvNCCheck can be found at: https://movidius.github.io/blog/mvNCCheck/ (scroll down to the debugging networks section).