Intel® Distribution of OpenVINO™ Toolkit
Community assistance about the Intel® Distribution of OpenVINO™ toolkit, OpenCV, and all aspects of computer vision-related on Intel® platforms.
6404 Discussions

Tiny YOLOv3 on NCS2 (FP16) in R5 SDK

nikos1
Valued Contributor I
‎12-21-2018 04:09 PM
1,868 Views

Anyone managed to get tiny YOLOv3 running on NCS2 in R5?

I am getting 

[ ERROR ] [VPU] Internal error: Output in 
detector/yolo-v3-tiny/pool2_5/MaxPool has incorrect width dimension. 
Expected: 9 or 9 Actual: 10

[ JFTR Non-tiny YOLOv3 now runs fine on NCS2 in the new R5 SDK and same tiny FP16 IR runs fine on GPU  ]

Thanks,

Nikos

0 Kudos

Link Copied

24 Replies
Ukhupacha
Beginner
‎02-15-2019 02:45 AM
228 Views

Hello!

Hyodo, Katsuya wrote:

@Stefano M.
https://github.com/PINTO0309/OpenVINO-YoloV3/blob/master/cpp/README.md

@Peniak, Martin
@Leini, Mikk
@om77
https://github.com/PINTO0309/OpenVINO-YoloV3/issues/7

 

Thanks for the implementation, I wonder how you came to the conclusion of using mask=3,4,5 because in the original yolo.cfg there are three [yolo] layers. [0,1,2] [3,4,5] and [6,78].

Same thing for yolov3-tiny

 

Kindly
Ukhupacha

0 Kudos
Copy link
Tsin__Ross
New Contributor I
‎03-18-2019 05:51 AM
228 Views

nikos wrote:

Sorry my mistake. Tiny YOLO v3 works fine in R5 SDK on NCS2  with FP16 IR ( size 416x416 ) .

Speed is about 20 fps - impressive!

 

performance counts:

LeakyReLU_                    OPTIMIZED_OUT  layerType: ReLU               realTime: 0          cpu: 0              execType: ReLU
LeakyReLU_837                 OPTIMIZED_OUT  layerType: ReLU               realTime: 0          cpu: 0              execType: ReLU
LeakyReLU_838                 OPTIMIZED_OUT  layerType: ReLU               realTime: 0          cpu: 0              execType: ReLU
LeakyReLU_838@soc=2/2@accum   EXECUTED       layerType: Convolution        realTime: 277        cpu: 277            execType: Sum
LeakyReLU_839                 OPTIMIZED_OUT  layerType: ReLU               realTime: 0          cpu: 0              execType: ReLU
LeakyReLU_840                 OPTIMIZED_OUT  layerType: ReLU               realTime: 0          cpu: 0              execType: ReLU
LeakyReLU_841                 OPTIMIZED_OUT  layerType: ReLU               realTime: 0          cpu: 0              execType: ReLU
LeakyReLU_842                 OPTIMIZED_OUT  layerType: ReLU               realTime: 0          cpu: 0              execType: ReLU
LeakyReLU_842 -> LeakyReLU... EXECUTED       layerType: Resample           realTime: 217        cpu: 217            execType: Permute
LeakyReLU_843                 OPTIMIZED_OUT  layerType: ReLU               realTime: 0          cpu: 0              execType: ReLU
LeakyReLU_844                 OPTIMIZED_OUT  layerType: ReLU               realTime: 0          cpu: 0              execType: ReLU
LeakyReLU_844@soc=2/2@accum   EXECUTED       layerType: Convolution        realTime: 261        cpu: 261            execType: Sum
LeakyReLU_845                 OPTIMIZED_OUT  layerType: ReLU               realTime: 0          cpu: 0              execType: ReLU
Receive-Tensor                EXECUTED       layerType: Receive-Tensor     realTime: 0          cpu: 0              execType: Receive-Tensor
detector/yolo-v3-tiny/Conv... EXECUTED       layerType: Convolution        realTime: 1774       cpu: 1774           execType: MyriadXHwConvolution
detector/yolo-v3-tiny/Conv... EXECUTED       layerType: Convolution        realTime: 1766       cpu: 1766           execType: MyriadXHwConvolution + injected[Copy]
detector/yolo-v3-tiny/Conv... EXECUTED       layerType: Convolution        realTime: 1770       cpu: 1770           execType: MyriadXHwConvolution + injected[Copy]
detector/yolo-v3-tiny/Conv... EXECUTED       layerType: Convolution        realTime: 1769       cpu: 1769           execType: MyriadXHwConvolution + injected[Copy]
detector/yolo-v3-tiny/Conv... EXECUTED       layerType: Convolution        realTime: 425        cpu: 425            execType: MyriadXHwConvolution + injected[Copy]
detector/yolo-v3-tiny/Conv... EXECUTED       layerType: Convolution        realTime: 22         cpu: 22             execType: Copy
detector/yolo-v3-tiny/Conv... EXECUTED       layerType: Convolution        realTime: 1408       cpu: 1408           execType: MyriadXHwConvolution
detector/yolo-v3-tiny/Conv... EXECUTED       layerType: Convolution        realTime: 1417       cpu: 1417           execType: MyriadXHwConvolution + injected[Copy]
detector/yolo-v3-tiny/Conv... EXECUTED       layerType: Convolution        realTime: 186        cpu: 186            execType: MyriadXHwConvolution + injected[Copy]
detector/yolo-v3-tiny/Conv... EXECUTED       layerType: Convolution        realTime: 18         cpu: 18             execType: Copy
detector/yolo-v3-tiny/Conv... EXECUTED       layerType: Convolution        realTime: 51         cpu: 51             execType: MyriadXHwConvolution
detector/yolo-v3-tiny/Conv... EXECUTED       layerType: Convolution        realTime: 2034       cpu: 2034           execType: MyriadXHwConvolution + injected[Permute]
detector/yolo-v3-tiny/Conv... EXECUTED       layerType: Convolution        realTime: 2092       cpu: 2092           execType: MyriadXHwConvolution
detector/yolo-v3-tiny/Conv... EXECUTED       layerType: Convolution        realTime: 203        cpu: 203            execType: LeakyRelu
detector/yolo-v3-tiny/Conv... EXECUTED       layerType: RegionYolo         realTime: 13435      cpu: 13435          execType: RegionYolo
detector/yolo-v3-tiny/Conv... EXECUTED       layerType: <Extra>            realTime: 266        cpu: 266            execType: Convert_f16f32
detector/yolo-v3-tiny/Conv... EXECUTED       layerType: Convolution        realTime: 325        cpu: 325            execType: MyriadXHwConvolution
detector/yolo-v3-tiny/Conv... EXECUTED       layerType: RegionYolo         realTime: 1278       cpu: 1278           execType: Permute
detector/yolo-v3-tiny/Conv... EXECUTED       layerType: Convolution        realTime: 1439       cpu: 1439           execType: MyriadXHwConvolution
detector/yolo-v3-tiny/Conv... EXECUTED       layerType: Convolution        realTime: 1290       cpu: 1290           execType: MyriadXHwConvolution
detector/yolo-v3-tiny/Conv... EXECUTED       layerType: Convolution        realTime: 1472       cpu: 1472           execType: MyriadXHwConvolution
detector/yolo-v3-tiny/Conv... EXECUTED       layerType: Convolution        realTime: 1743       cpu: 1743           execType: MyriadXHwConvolution
detector/yolo-v3-tiny/Conv... EXECUTED       layerType: Convolution        realTime: 3417       cpu: 3417           execType: MyriadXHwConvolution
detector/yolo-v3-tiny/Conv... EXECUTED       layerType: Convolution        realTime: 3352       cpu: 3352           execType: MyriadXHwConvolution
detector/yolo-v3-tiny/Conv... EXECUTED       layerType: Convolution        realTime: 205        cpu: 205            execType: LeakyRelu
detector/yolo-v3-tiny/Conv... EXECUTED       layerType: Convolution        realTime: 336        cpu: 336            execType: MyriadXHwConvolution
detector/yolo-v3-tiny/Resi... EXECUTED       layerType: Resample           realTime: 121        cpu: 121            execType: Resample
detector/yolo-v3-tiny/conc... OPTIMIZED_OUT  layerType: Concat             realTime: 0          cpu: 0              execType: Concat
detector/yolo-v3-tiny/conc... EXECUTED       layerType: Convolution        realTime: 977        cpu: 977            execType: Permute
detector/yolo-v3-tiny/pool... OPTIMIZED_OUT  layerType: Pooling            realTime: 0          cpu: 0              execType: Pooling
detector/yolo-v3-tiny/pool... OPTIMIZED_OUT  layerType: Pooling            realTime: 0          cpu: 0              execType: Pooling
detector/yolo-v3-tiny/pool... OPTIMIZED_OUT  layerType: Pooling            realTime: 0          cpu: 0              execType: Pooling
detector/yolo-v3-tiny/pool... OPTIMIZED_OUT  layerType: Pooling            realTime: 0          cpu: 0              execType: Pooling
detector/yolo-v3-tiny/pool... EXECUTED       layerType: Pooling            realTime: 457        cpu: 457            execType: MyriadXHwPooling + injected[Permute]
detector/yolo-v3-tiny/pool... EXECUTED       layerType: Pooling            realTime: 140        cpu: 140            execType: MyriadXHwPooling
detector/yolo-v3-tiny/pool... EXECUTED       layerType: Pooling            realTime: 1242       cpu: 1242           execType: CopyMakeBorder
inputs@FP16                   EXECUTED       layerType: <Extra>            realTime: 474        cpu: 474            execType: Convert_u8f16
Total time: 47659    microseconds
[ WARN:0] terminating async callback
[ INFO ] Execution successful

 

 

How do you do that?

I updated my OpenVINO to the latest version, 

when converting .pb model to IR model, I got this,

Model Optimizer version:        1.5.12.49d067a0
[ ERROR ]  List of operations that cannot be converted to IE IR:
[ ERROR ]      LeakyRelu (11)
[ ERROR ]          detector/yolo-v3-tiny/Conv/LeakyRelu
[ ERROR ]          detector/yolo-v3-tiny/Conv_1/LeakyRelu
[ ERROR ]          detector/yolo-v3-tiny/Conv_2/LeakyRelu
[ ERROR ]          detector/yolo-v3-tiny/Conv_3/LeakyRelu
[ ERROR ]          detector/yolo-v3-tiny/Conv_4/LeakyRelu
[ ERROR ]          detector/yolo-v3-tiny/Conv_5/LeakyRelu
[ ERROR ]          detector/yolo-v3-tiny/Conv_6/LeakyRelu
[ ERROR ]          detector/yolo-v3-tiny/Conv_7/LeakyRelu
[ ERROR ]          detector/yolo-v3-tiny/Conv_10/LeakyRelu
[ ERROR ]          detector/yolo-v3-tiny/Conv_11/LeakyRelu
[ ERROR ]          detector/yolo-v3-tiny/Conv_8/LeakyRelu
[ ERROR ]  Part of the nodes was not translated to IE. Stopped.
 For more information please refer to Model Optimizer FAQ (<INSTALL_DIR>/deployment_tools/documentation/docs/MO_FAQ.html), question #24.

 

 

0 Kudos
Copy link
Hyodo__Katsuya
Innovator
‎03-18-2019 09:53 AM
228 Views

@Tsin, Ross

Please try below.

sudo -H pip3 install tensorflow==1.12.0 --upgrade

or

sudo -H pip3 install tensorflow-gpu==1.12.0 --upgrade

 

0 Kudos
Copy link
RTasa
New Contributor I
‎03-19-2019 09:43 AM
228 Views
There are a lot or new accelerators out there. The NCS2 is old news in a way with the announcement of Google Coral https://coral.withgoogle.com/ NVIDIA Jetson Nano https://developer.nvidia.com/embedded/buy/jetson-nano-devkit Anyone privy to an insider track on these? Pros cons etc.
0 Kudos
Copy link
Reply

Community support is provided during standard business hours (Monday to Friday 7AM - 5PM PST). Other contact methods are available here.

Intel does not verify all solutions, including but not limited to any file transfers that may appear in this community. Accordingly, Intel disclaims all express and implied warranties, including without limitation, the implied warranties of merchantability, fitness for a particular purpose, and non-infringement, as well as any warranty arising from course of performance, course of dealing, or usage in trade.

For more complete information about compiler optimizations, see our Optimization Notice.