Inference engine error dependent on image input size

Inference engine error dependent on image input size

I'm using OpenVINO R4 on NCS2. I have a 2 layer convolutional neural network, with the following topology:
(20, 5, 2, 2), (32, 3, 1, 1). The tuples are encoded as follows: (num_channels, kernel_size, padding, stride).

As a first test, I'm trying to pass dummy data through this network. With an input size of (1, 3, 128, 128*11), the data is passed through the network just fine. However, with an input size of  (1, 3, 128, 128*10), I get the following error:

E: [xLink] [         0] dispatcherEventReceive:308      dispatcherEventReceive() Read failed -4 | event 0x7f0aee7fbeb0 USB_READ_REL_RESP

E: [xLink] [         0] eventReader:254 eventReader stopped

Moreover, while the input size (1, 3, 128, 128*11) works fine with the network topology that I mentioned, it again fails with the following network configuration:

(20, 5, 2, 2),
(32, 3, 1, 1),
(40, 1, 0, 1), 
(40, 3, 1, 2),
(48, 3, 1, 1), 
(60, 1, 0, 1)
 

 

1 post / 0 new
For more complete information about compiler optimizations, see our Optimization Notice.