OpenVINO-YoloV3

YoloV3 / tiny-YoloV3 + RaspberryPi3 / Ubuntu LaptopPC + NCS/NCS2 + USB Camera + Python

Inspired from https://github.com/mystic123/tensorflow-yolo-v3.git

Performance comparison as a mobile application (Based on sensory comparison)
◯=HIGH, △=MEDIUM, ×=LOW

No.	Model	Speed	Accuracy	Adaptive distance
1	SSD	×	◯	ALL
2	MobileNet-SSD	△	△	Short distance
3	YoloV3	×	◯	ALL
4	tiny-YoloV3	◯	△	Long distance

My articles

[24 FPS] Boost RaspberryPi3 with four Neural Compute Stick 2 (NCS2) MobileNet-SSD / YoloV3 [48 FPS for Core i7]
[13 FPS] NCS2 x 4 + Full size YoloV3 performance has been tripled
Support for local training and OpenVINO of One Class tiny-YoloV3 with a proprietary data set

Change history

[Mar 01, 2019] Improve accuracy. Fixed preprocessing and postprocessing bug.
[Mar 17, 2019] Added a training procedure with your own data set.
[Apr 03, 2019] Work on OpenVINO 2019 R1 started.
[Apr 14, 2019] Compatible with 2019 R1.
[Apr 26, 2019] Compatible with 2019 R1.0.1.

Operation sample

<CPP + YoloV3 - Intel Core i7-8750H, CPU Only, 4 FPS - 5 FPS>

<CPP + tiny-YoloV3 - Intel Core i7-8750H, CPU Only, 60 FPS>

<Python + tiny-YoloV3 + USBCamera, Core i7-8750H, CPU Only, 30 FPS>

<Python + tiny-YoloV3 + Async + USBCamera, Core i7-8750H, NCS2, 30 FPS+>
To raise the detection rate, lower the threshold by yourself.
The default threshold is 40%.

<Python + YoloV3 + MP4, Core i7-8750H, NCS2 x4, 13 FPS>
【Note】 Due to the performance difference of ARM <-> Core series, performance is degraded in RaspberryPi3.

Python Version YoloV3 / tiny-YoloV3 (Dec 28, 2018 Operation confirmed)

YoloV3

$ python3 openvino_yolov3_test.py

tiny-YoloV3 + NCS2 MultiStick

$ python3 openvino_tiny-yolov3_MultiStick_test.py -numncs 1

YoloV3 + NCS2 MultiStick (Pretty slow)

$ python3 openvino_yolov3_MultiStick_test.py -numncs 4

CPP Version YoloV3 / tiny-YoloV3 (Dec 16, 2018 Operation confirmed)

cpp version is here "cpp/object_detection_demo_yolov3_async"

Environment

LattePanda Alpha (Intel 7th Core m3-7y30) or LaptopPC (Intel 8th Core i7-8750H)
Ubuntu 16.04 x86_64
RaspberryPi3
Raspbian Stretch armv7l
OpenVINO toolkit 2019 R1.0.1 (2019.1.133)
Python 3.5
OpenCV 4.1.0-openvino
Tensorflow v1.12.0 or Tensorflow-GPU v1.12.0 (pip install)
YoloV3 (MS-COCO)
tiny-YoloV3 (MS-COCO)
USB Camera (PlaystationEye) / Movie file (mp4)
Intel Neural Compute Stick v1 / v2

OpenVINO Supported Layers (As of Apr 14, 2019)

Supported Devices (https://docs.openvinotoolkit.org/latest/_docs_IE_DG_supported_plugins_Supported_Devices.html#supported_layers)

Layers	GPU	CPU	MYRIAD(VPU)	GNA	FPGA	ShapeInfer
Activation-Clamp	Supported	Supported	Supported	Supported	Supported	Supported
Activation-ELU	Supported	Supported	Supported	Not Supported	Supported	Supported
Activation-Leaky ReLU	Supported	Supported	Supported	Supported	Supported	Supported
Activation-PReLU	Supported	Supported	Supported	Not Supported	Supported	Supported
Activation-ReLU	Supported	Supported	Supported	Supported	Supported	Supported
Activation-ReLU6	Supported	Supported	Supported	Not Supported	Not Supported	Supported
Activation-Sigmoid/Logistic	Supported	Supported	Supported	Supported	Not Supported	Supported
Activation-TanH	Supported	Supported	Supported	Supported	Not Supported	Supported
ArgMax	Supported	Supported	Supported	Not Supported	Not Supported	Supported
BatchNormalization	Supported	Supported	Supported	Not Supported	Supported	Supported
Concat	Supported	Supported	Supported	Supported	Supported	Supported
Const	Supported	Supported	Supported	Not Supported	Not Supported	Not Supported
Convolution-Dilated	Supported	Supported	Supported	Not Supported	Not Supported	Supported
Convolution-Dilated 3D	Not Supported	Supported	Not Supported	Not Supported	Not Supported	Not Supported
Convolution-Grouped	Supported	Supported	Supported	Not Supported	Supported	Supported
Convolution-Grouped 3D	Not Supported	Supported	Not Supported	Not Supported	Not Supported	Not Supported
Convolution-Ordinary	Supported	Supported	Supported	Supported	Supported	Supported
Convolution-Ordinary 3D	Not Supported	Supported	Not Supported	Not Supported	Not Supported	Not Supported
Crop	Supported	Supported	Supported	Supported	Not Supported	Supported
CTCGreedyDecoder	Supported	Supported	Supported	Not Supported	Not Supported	Supported
Deconvolution	Supported	Supported	Supported	Not Supported	Supported	Supported
Deconvolution 3D	Not Supported	Supported	Not Supported	Not Supported	Not Supported	Not Supported
DetectionOutput	Supported	Supported	Supported	Not Supported	Not Supported	Supported
Eltwise-Max	Supported	Supported	Supported	Not Supported	Not Supported	Supported
Eltwise-Mul	Supported	Supported	Supported	Supported	Not Supported	Supported
Eltwise-Sum	Supported	Supported	Supported	Supported	Supported	Supported
Flatten	Supported	Supported	Supported	Not Supported	Not Supported	Supported
FullyConnected (Inner Product)	Supported	Supported	Supported	Supported	Supported	Supported
Gather	Not Supported	Supported	Not Supported	Not Supported	Not Supported	Supported
Gemm	Not Supported	Supported	Not Supported	Not Supported	Not Supported	Supported
GRN	Supported	Supported	Supported	Not Supported	Not Supported	Supported
Interp	Supported	Supported	Supported	Not Supported	Not Supported	Supported
LRN (Norm)	Supported	Supported	Supported	Not Supported	Supported	Supported
LSTMCell	Supported	Supported	Not Supported	Not Supported	Not Supported	Not Supported
GRUCell	Supported	Supported	Not Supported	Not Supported	Not Supported	Not Supported
RNNCell	Supported	Supported	Not Supported	Not Supported	Not Supported	Not Supported
LSTMSequence	Supported	Supported	Not Supported	Not Supported	Not Supported	Not Supported
GRUSequence	Supported	Supported	Not Supported	Not Supported	Not Supported	Not Supported
RNNSequence	Supported	Supported	Not Supported	Not Supported	Not Supported	Not Supported
Memory	Not Supported	Supported	Not Supported	Supported	Not Supported	Supported
MVN	Supported	Supported	Supported	Not Supported	Not Supported	Supported
Normalize	Supported	Supported	Supported	Not Supported	Not Supported	Supported
Pad	Supported	Supported	Supported	Not Supported	Not Supported	Supported
Permute	Supported	Supported	Supported	Supported	Not Supported	Supported
Pooling(AVG,MAX)	Supported	Supported	Supported	Supported	Supported	Supported
Pooling(AVG,MAX) 3D	Not Supported	Supported	Not Supported	Not Supported	Not Supported	Not Supported
Power	Supported	Supported	Supported	Supported	Supported	Supported
PriorBox	Supported	Supported	Supported	Not Supported	Not Supported	Supported
PriorBoxClustered	Supported	Supported	Supported	Not Supported	Not Supported	Supported
Proposal	Supported	Supported	Supported	Not Supported	Not Supported	Supported
PSROIPooling	Supported	Supported	Supported	Not Supported	Not Supported	Supported
RegionYolo	Supported	Supported	Supported	Not Supported	Not Supported	Supported
ReorgYolo	Supported	Supported	Supported	Not Supported	Not Supported	Supported
Resample	Supported	Supported	Supported	Not Supported	Not Supported	Supported
Reshape	Supported	Supported	Supported	Supported	Not Supported	Supported
RNN	Not Supported	Supported	Supported	Not Supported	Not Supported	Not Supported
ROIPooling	Supported	Supported	Supported	Not Supported	Not Supported	Supported
ScaleShift	Supported	Supported	Supported	Supported	Supported	Supported
SimplerNMS	Supported	Supported	Not Supported	Not Supported	Not Supported	Supported
Slice	Supported	Supported	Supported	Supported	Supported	Supported
SoftMax	Supported	Supported	Supported	Not Supported	Not Supported	Supported
SpatialTransformer	Not Supported	Supported	Not Supported	Not Supported	Not Supported	Supported
Split	Supported	Supported	Supported	Supported	Supported	Supported
TensorIterator	Not Supported	Supported	Not Supported	Not Supported	Not Supported	Not Supported
Tile	Supported	Supported	Supported	Not Supported	Not Supported	Supported
Unpooling	Supported	Not Supported	Not Supported	Not Supported	Not Supported	Not Supported
Upsampling	Supported	Not Supported	Not Supported	Not Supported	Not Supported	Not Supported

OpenVINO - Python API

https://docs.openvinotoolkit.org/latest/_inference_engine_ie_bridges_python_docs_api_overview.html

Environment construction procedure

1. Work with LaptopPC (Ubuntu 16.04)

1.OpenVINO 2019R1.0.1 Full-Install. Execute the following command.

$ cd ~
$ curl -sc /tmp/cookie "https://drive.google.com/uc?export=download&id=1ciX7cHqCh8lLFYI0HKkhC3r_fMirrlKk" > /dev/null
$ CODE="$(awk '/_warning_/ {print $NF}' /tmp/cookie)"
$ curl -Lb /tmp/cookie "https://drive.google.com/uc?export=download&confirm=${CODE}&id=1ciX7cHqCh8lLFYI0HKkhC3r_fMirrlKk" -o l_openvino_toolkit_p_2019.1.133.tgz
$ tar -zxf l_openvino_toolkit_p_2019.1.133.tgz
$ rm l_openvino_toolkit_p_2019.1.133.tgz
$ cd l_openvino_toolkit_p_2019.1.133
$ sudo -E ./install_openvino_dependencies.sh

## GUI version installer
$ sudo ./install_GUI.sh
 or
## CUI version installer
$ sudo ./install.sh

2.Configure the Model Optimizer. Execute the following command.

$ cd /opt/intel/openvino/install_dependencies/
$ sudo -E ./install_openvino_dependencies.sh
$ nano ~/.bashrc
source /opt/intel/openvino/bin/setupvars.sh

$ source ~/.bashrc
$ cd /opt/intel/openvino/deployment_tools/model_optimizer/install_prerequisites/
$ sudo ./install_prerequisites.sh

3.【Optional execution】 Additional installation steps for the Intel® Movidius™ Neural Compute Stick v1 and Intel® Neural Compute Stick v2

$ sudo usermod -a -G users "$(whoami)"
$ cat <<EOF > 97-usbboot.rules
SUBSYSTEM=="usb", ATTRS{idProduct}=="2150", ATTRS{idVendor}=="03e7", GROUP="users", MODE="0666", ENV{ID_MM_DEVICE_IGNORE}="1"
SUBSYSTEM=="usb", ATTRS{idProduct}=="2485", ATTRS{idVendor}=="03e7", GROUP="users", MODE="0666", ENV{ID_MM_DEVICE_IGNORE}="1"
SUBSYSTEM=="usb", ATTRS{idProduct}=="f63b", ATTRS{idVendor}=="03e7", GROUP="users", MODE="0666", ENV{ID_MM_DEVICE_IGNORE}="1"
EOF

$ sudo cp 97-usbboot.rules /etc/udev/rules.d/
$ sudo udevadm control --reload-rules
$ sudo udevadm trigger
$ sudo ldconfig
$ rm 97-usbboot.rules

4.【Optional execution】 Additional installation steps for processor graphics (GPU, Intel HD Graphics series only)

$ cd /opt/intel/openvino/install_dependencies/
$ sudo -E su
$ uname -r
4.15.0-42-generic #<--- display kernel version sample

### Execute only when the kernel version is older than 4.14
$ ./install_4_14_kernel.sh

$ ./install_NEO_OCL_driver.sh
$ sudo reboot

2. Work with RaspberryPi (Raspbian Stretch)

[Note] Only the execution environment is introduced.

1.Execute the following command.

$ sudo apt update
$ sudo apt upgrade
$ curl -sc /tmp/cookie "https://drive.google.com/uc?export=download&id=1NFt6g6ZkneHioU2P7rUJ8BFpQhIazbym" > /dev/null
$ CODE="$(awk '/_warning_/ {print $NF}' /tmp/cookie)"
$ curl -Lb /tmp/cookie "https://drive.google.com/uc?export=download&confirm=${CODE}&id=1NFt6g6ZkneHioU2P7rUJ8BFpQhIazbym" -o l_openvino_toolkit_raspbi_p_2019.1.133.tgz
$ tar -zxvf l_openvino_toolkit_raspbi_p_2019.1.133.tgz
$ rm l_openvino_toolkit_raspbi_p_2019.1.133.tgz
$ sed -i "s|<INSTALLDIR>|$(pwd)/inference_engine_vpu_arm|" inference_engine_vpu_arm/bin/setupvars.sh

2.Execute the following command.

$ nano ~/.bashrc
### Add 1 row below
source /home/pi/inference_engine_vpu_arm/bin/setupvars.sh

$ source ~/.bashrc
### Successful if displayed as below
[setupvars.sh] OpenVINO environment initialized

$ sudo usermod -a -G users "$(whoami)"
$ sudo reboot

3.Update USB rule.

$ sh inference_engine_vpu_arm/install_dependencies/install_NCS_udev_rules.sh
### It is displayed as follows
Update udev rules so that the toolkit can communicate with your neural compute stick
[install_NCS_udev_rules.sh] udev rules installed

[Note] OpenCV 4.1.0 will be installed without permission when the work is finished. If you do not want to affect other environments, please edit environment variables after installation is completed.

Training with your own data set

See the article below.
A sample of one-class training with Darknet and tiny-YoloV3.
https://qiita.com/PINTO/items/7dd7135085a7249bf17a#support-for-local-training-and-openvino-of-one-class-tiny-yolov3-with-a-proprietary-data-set

How to install Bazel (version 0.17.2, x86_64 only)

1. Bazel introduction command

$ cd ~
$ curl -sc /tmp/cookie "https://drive.google.com/uc?export=download&id=1dvR3pdM6vtkTWqeR-DpgVUoDV0EYWil5" > /dev/null
$ CODE="$(awk '/_warning_/ {print $NF}' /tmp/cookie)"
$ curl -Lb /tmp/cookie "https://drive.google.com/uc?export=download&confirm=${CODE}&id=1dvR3pdM6vtkTWqeR-DpgVUoDV0EYWil5" -o bazel
$ sudo cp ./bazel /usr/local/bin
$ rm ./bazel

2. Supplementary information

https://github.com/PINTO0309/Bazel_bin.git

How to check the graph structure of a ".pb" file [Part.1]

Simple structure analysis.

1. Build and run graph structure analysis program

$ cd ~
$ git clone -b v1.11.0 https://github.com/tensorflow/tensorflow.git
$ cd tensorflow
$ git checkout -b v1.11.0
$ bazel build tensorflow/tools/graph_transforms:summarize_graph
$ bazel-bin/tensorflow/tools/graph_transforms/summarize_graph --in_graph=xxxx.pb

2. Sample of display result

YoloV3

Found 1 possible inputs: (name=inputs, type=float(1), shape=[?,416,416,3]) 
No variables spotted.
Found 1 possible outputs: (name=output_boxes, op=ConcatV2) 
Found 62002034 (62.00M) const parameters, 0 (0) variable parameters, and 0 control_edges
Op types used: 536 Const, 372 Identity, 87 Mul, 75 Conv2D, 72 FusedBatchNorm, 72 Maximum, 28 Add, \
24 Reshape, 14 ConcatV2, 9 Sigmoid, 6 Tile, 6 Range, 5 Pad, 4 SplitV, 3 Pack, 3 RealDiv, 3 Fill, \
3 Exp, 3 BiasAdd, 2 ResizeNearestNeighbor, 2 Sub, 1 Placeholder
To use with tensorflow/tools/benchmark:benchmark_model try these arguments:
bazel run tensorflow/tools/benchmark:benchmark_model -- \
--graph=/home/b920405/git/OpenVINO-YoloV3/pbmodels/frozen_yolo_v3.pb \
--show_flops \
--input_layer=inputs \
--input_layer_type=float \
--input_layer_shape=-1,416,416,3 \
--output_layer=output_boxes

tiny-YoloV3

Found 1 possible inputs: (name=inputs, type=float(1), shape=[?,416,416,3]) 
No variables spotted.
Found 1 possible outputs: (name=output_boxes, op=ConcatV2) 
Found 8858858 (8.86M) const parameters, 0 (0) variable parameters, and 0 control_edges
Op types used: 134 Const, 63 Identity, 21 Mul, 16 Reshape, 13 Conv2D, 11 FusedBatchNorm, 11 Maximum, \
10 ConcatV2, 6 Sigmoid, 6 MaxPool, 4 Tile, 4 Add, 4 Range, 3 RealDiv, 3 SplitV, 2 Pack, 2 Fill, \
2 Exp, 2 Sub, 2 BiasAdd, 1 Placeholder, 1 ResizeNearestNeighbor
To use with tensorflow/tools/benchmark:benchmark_model try these arguments:
bazel run tensorflow/tools/benchmark:benchmark_model -- \
--graph=/home/b920405/git/OpenVINO-YoloV3/pbmodels/frozen_tiny_yolo_v3.pb \
--show_flops \
--input_layer=inputs \
--input_layer_type=float \
--input_layer_shape=-1,416,416,3 \
--output_layer=output_boxes

How to check the graph structure of a ".pb" file [Part.2]

Convert to text format.

1. Run graph structure analysis program

$ python3 tfconverter.py
### ".pbtxt" in ProtocolBuffer format is output.
### The size of the generated text file is huge.

How to check the graph structure of a ".pb" file [Part.3]

Use Tensorboard.

1. Run log output program for Tensorboard

import tensorflow as tf
from tensorflow.python.platform import gfile

with tf.Session() as sess:
    model_filename ="xxxx.pb"
    with gfile.FastGFile(model_filename, "rb") as f:
        graph_def = tf.GraphDef()
        graph_def.ParseFromString(f.read())
        g_in = tf.import_graph_def(graph_def)

    LOGDIR="path/to/logs"
    train_writer = tf.summary.FileWriter(LOGDIR)
    train_writer.add_graph(sess.graph)

2. Starting Tensorboard

$ tensorboard --logdir=path/to/logs

3. Display of Tensorboard

Access http://localhost:6006 from the browser.

How to check the graph structure of a ".pb" file [Part.4]

Use netron.

1. Install netron

$ sudo -H pip3 install netron

2. Starting netron

$ netron -b [MODEL_FILE]

3. Display of netron

Access http://localhost:8080 from the browser.

Neural Compute Stick 2

https://ncsforum.movidius.com/discussion/1302/intel-neural-compute-stick-2-information

Issue

OpenVINO failing on YoloV3's YoloRegion, only one working on FP16, all working on FP32
Regarding YOLO family networks on NCS2. Possibly a work-around
Convert YOLOv3 Model to IR

Reference

https://github.com/opencv/opencv/wiki/Intel%27s-Deep-Learning-Inference-Engine-backend https://github.com/opencv/opencv/wiki/Intel%27s-Deep-Learning-Inference-Engine-backend#raspbian-stretch

jeonggunlee/OpenVINO-YoloV3

OpenVINO-YoloV3

My articles

Change history

Operation sample

Python Version YoloV3 / tiny-YoloV3 (Dec 28, 2018 Operation confirmed)

YoloV3

tiny-YoloV3 + NCS2 MultiStick

YoloV3 + NCS2 MultiStick (Pretty slow)

CPP Version YoloV3 / tiny-YoloV3 (Dec 16, 2018 Operation confirmed)

Environment

OpenVINO Supported Layers (As of Apr 14, 2019)

OpenVINO - Python API

Environment construction procedure

1. Work with LaptopPC (Ubuntu 16.04)

2. Work with RaspberryPi (Raspbian Stretch)

Training with your own data set

How to install Bazel (version 0.17.2, x86_64 only)

1. Bazel introduction command

2. Supplementary information

How to check the graph structure of a ".pb" file [Part.1]

1. Build and run graph structure analysis program

2. Sample of display result

How to check the graph structure of a ".pb" file [Part.2]

1. Run graph structure analysis program

How to check the graph structure of a ".pb" file [Part.3]

1. Run log output program for Tensorboard

2. Starting Tensorboard

3. Display of Tensorboard

How to check the graph structure of a ".pb" file [Part.4]

1. Install netron

2. Starting netron

3. Display of netron

Neural Compute Stick 2

Issue

Reference