Developing Games & Graphics on Intel
Ask us questions and get answers, tips, tweaks & techniques from product and technology experts.
Announcements
This is a Peer-to-Peer forum only. We apologize for any inconvenience.

Learn more about developing on Intel® Hardware and Software here.
439 Discussions

NCS2 does not generate correct conv results when strides set to 2

Tsin__Ross
New Contributor I
186 Views

Hi, 

I write my IR xml file following the docs in https://docs.openvinotoolkit.org/latest/openvino_docs_MO_DG_prepare_model_convert_model_Legacy_IR_La... 

and the first layer look like this:

<net batch = "1" name="shufflenet-myriad" version="7">
	<layers>
		<layer id="0" name="inputs" precision="FP32" type="Input" >
			<output>
				<port id="0" >
					<dim>1</dim>
					<dim>3</dim>
					<dim>416</dim>
					<dim>416</dim>
				</port>
			</output>
		</layer>
		<layer id="1" name="layer01.conv1" precision="FP16" type="Convolution" >
			<data auto_pad="same_upper" dilations="1,1" group="1" kernel="3,3" output="24" pads_begin="1,1" pads_end="1,1" strides="2,2"/>
			<weights offset="0" size="1296"/>
			<biases offset="1296" size="48"/>
			<input>
				<port id="0" >
					<dim>1</dim>
					<dim>3</dim>
					<dim>416</dim>
					<dim>416</dim>
				</port>
			</input>
			<output>
				<port id="1" >
					<dim>1</dim>
					<dim>24</dim>
					<dim>208</dim>
					<dim>208</dim>
				</port>
			</output>
		</layer>
	</layers>
	<edges>
		<edge from-layer="0" from-port="0" to-layer="1" to-port="0"/>
	</edges>
	<meta_data>
	</meta_data>
</net>

 

I did not get results as the GPU did, and checked the results, the differences were everywhere.cmp.jpg

Input data(a normalized 3x416x416 image file data ), weight, bias and expected output files are attached. 

Please help me to see whether this is a computation problem.

 

Thanks and Regards,

 

Ross

0 Kudos
1 Reply
Tsin__Ross
New Contributor I
174 Views

I removed the "auto_pad" attr and it worked.

Reply