Monday, September 12, 2016

[Haar Classifier] Train your own OpenCV Haar classifier

I just keep a record for myself because there are a lot of documents teaching how to train your haar classifier and almost of them seem to don't work well. The following 2 items are clear and easy to understand.

The Data Image Source (cars) I use.

1. Train your own OpenCV Haar classifier

find ./positive_images -iname "*.pgm" > positives.txt
find ./negative_images -iname "*.pgm" > negatives.txt

perl bin/ positives.txt negatives.txt samples 550\
  "opencv_createsamples -bgcolor 0 -bgthresh 0 -maxxangle 1.1\
  -maxyangle 1.1 maxzangle 0.5 -maxidev 40 -w 48 -h 24"

python ./tools/ -v samples/ -o samples.vec

opencv_traincascade -data classifier -vec samples.vec -bg negatives.txt\
  -numStages 10 -minHitRate 0.999 -maxFalseAlarmRate 0.5 -numPos 1000\
  -numNeg 600 -w 48 -h 24 -mode ALL -precalcValBufSize 1024\
  -precalcIdxBufSize 1024

2. OpenCV Tutorial: Training your own detector |
find pos/ -name '*.pgm' -exec echo \{\} 1 0 0 100 40 \; >
find neg/ -name '*.pgm' > bg.txt

opencv_createsamples -info -num 550 -w 48 -h 24 -vec cars.vec
opencv_createsamples -w 48 -h 24 -vec cars.vec

opencv_traincascade -data data -vec cars.vec -bg bg.txt \
 -numPos 500 -numNeg 500 -numStages 10 -w 48 -h 24 -featureType LBP

P.S: Which one is best? I don't know...

[Image] How to resize, convert & modify images from the Linux

$ sudo apt-get install imagemagick

Converting Between Formats
$ convert howtogeek.png howtogeek.jpg

You can also specify a compression level for JPEG images:
$ convert howtogeek.png -quality 95 howtogeek.jpg

Resizing Images
$ convert example.png -resize 200×100 example.png
  - to force the image to become a specific size – even if it messes up the aspect ratio
$ convert example.png -resize 200×100! example.png

$ convert example.png -resize 200 example.png
$ convert example.png -resize x100 example.png

Rotating an Image
convert howtogeek.jpg -rotate 90 howtogeek-rotated.jpg

Applying Effects
ImageMagick can apply a variety of effects to an image.
  - For example,   the following command applies the “charcoal” effect to an image:
$ convert howtogeek.jpg -charcoal 2 howtogeek-charcoal.jpg
  - the “Implode” effect with a strength of 1:
# convert howtogeek.jpg -implode 1 howtogeek-imploded.jpg

Batch Processing
for file in *.png; do convert $file -rotate 90 rotated-$file; done


Thursday, September 8, 2016

[TensorFlow] My case to install TensorFlow with GPU enabled

My Operation System is Ubuntu 14.04 LTS 5 and GPU card is GeForce GTX 750Ti

1. Go to and download the driver (
2. For Nvidia to find linux header files (*):
$ sudo apt-get install build-essential linux-headers-$(uname -r)

3. To enable full screen text mode (nomodeset):
$ sudo gedit /etc/default/grub
>> Edit GRUB_CMDLINE_LINUX_DEFAULT="quiet splash nomodeset"
Save it and reboot
$ sudo update-grub
$ sudo reboot

4. Log into with Ctl +Alt + F1

5. Stop the X Server service
$ sudo service lightdm stop

6. Install nVidia driver
$ sudo ./
7. Install CUDA (GPUs on Linux)
Download and install Cuda Toolkit dpkg -i cuda-repo-ubuntu1404-7-5-local_7.5-18_amd64.deb
sudo apt-get update
sudo apt-get install cuda

8. Download and install cuDNN xvzf cudnn-7.0-linux-x64-v4.tgz
sudo cp cudnn-7.0-linux-x64-v4/cudnn.h /usr/local/cuda/include
sudo cp cudnn-7.0-linux-x64-v4/libcudnn* /usr/local/cuda/lib64
sudo chmod a+r /usr/local/cuda/lib64/libcudnn*
9. You also need to set the LD_LIBRARY_PATH and CUDA_HOME environment variables. Consider adding the commands below to your ~/.bash_profile. These assume your CUDA installation is in /usr/local/cuda:
$ vim ~/.bashrc
>> export LD_LIBRARY_PATH="$LD_LIBRARY_PATH:/usr/local/cuda/lib64"
>> export CUDA_HOME=/usr/local/cuda
>> export PATH="$CUDA_HOME/bin:%PATH"
>> export PATH="$PATH:$HOME/bin" 
10. To install TensorFlow for Ubuntu/Linux 64-bit, GPU enabled:
$ sudo pip install --upgrade
To find out which device is used, you can enable log device placement like this:
$ python
>>>> import tensorflow as tf
>>>> sess = tf.Session(config=tf.ConfigProto(log_device_placement=True))

Tuesday, August 16, 2016

[OpenCV] To install OpenCV on Debian and create a test project using Netbeans

This document is the steps to install OpenCV on Debian and create a test project using Netbeans for my reference in case.

#Prepare the build environment GCC、Cmake、pkgconfig
$sudo apt-get -y install build-essential cmake pkg-config

#Install ImageI/O libraries
$sudo apt-get -y install libjpeg62-dev libtiff4-dev libjasper-dev

#Install Viode I/O libraries
$sudo apt-get -y install libavcodec-dev libavformat-dev libswscale-dev libv4l-dev
$sudo apt-get -y install libdc1394-22-dev libxine-dev libgstreamer0.10-dev libgstreamer-plugins-base0.10-dev

#Install GTK+2.x and QT libraries
$sudo apt-get -y install  libgtk2.0-dev libqt4-dev

#(Optional)Install support for Firewire video cameras
$sudo apt-get -y install libdc1394-22-dev

#(Optional)Install video streaming libraries
$sudo apt-get -y install libxine-dev libgstreamer0.10-dev libgstreamer-plugins-base0.10-dev

#(Optional)Install the Python development environment and the Python Numerical libraries
$sudo apt-get -y install python-dev python-numpy

#(Optional)Install TBB library(Intel® Threading Building Blocks)
$sudo apt-get -y install libtbb-dev

#Install libraries that are used to optimize various routines inside of OpenCV:
$sudo apt-get install libatlas-base-dev gfortran

#Download OpenCV and unpack it:
$wget -O
$wget -O
$cd opencv-2.4.13

#mkdir build and make OpenCV
$mkdir build
$cd build
$make -j4
$sudo make install
$sudo ldconfig

I use Netbeans as my IDE tool to develop and program OpenCV GUI and applications.


Thursday, August 11, 2016

[Hadoop] Setting up a Single Node Cluster

Basically these resource links are good enough to do a single node cluster of Hadoop MapReduce. But I still want to add some comments for my reference.

Login the user "hadoop"
# sudo su - hadoop

Go to the location of Hadoop
# /usr/local/hadoop

Add the variables in ~/.bashrc
export JAVA_HOME=/usr/lib/jvm/java-7-openjdk-amd64
export HADOOP_HOME=/usr/local/hadoop
export HADOOP_CONF_DIR=$HADOOP_HOME/etc/hadoop
export HADOOP_INSTALL=/usr/local/hadoop

Modify $JAVA_HOME in etc/hadoop/
export JAVA_HOME=/usr/lib/jvm/java-7-openjdk-amd64

Start dfs and yarn
# sbin/
# sbin/

Finally, we can try the Hadoop MapReduce example as follows:
# bin/hadoop jar share/hadoop/mapreduce/hadoop-mapreduce-examples-2.7.0.jar grep input output 'dfs[a-z.]+'

In order to forcefully let the namenode leave safemode, following command should be executed:
# hdfs dfsadmin -safemode leave

Monday, August 1, 2016

[Tensorflow] Fizz-Buzz example enhancement

I am just based on this Fizz-Buzz example as below to add 2nd convolution layer and guess what? The result is quicker to be learn. But, this is just the first step to learn "Deep Learning"...
There is still a lot of things and knowledge that need to learn more.




Sunday, July 24, 2016

[Neutron] The first glance of L3HA mode in OpenStack Neutron ( Liberty version )

I just quickly take the first glance of L3HA mode in OpenStack Neutron ( Liberty version ) and is based on my tenant environment as follows:

My tenant environment

# neutron router-list

# neutron net-list

# neutron subnet-list

The Topology view looks like this:

Here I have 2 instances in my tenant:

So, if I use the instance: daanny_vm1 to ping danny_vm2, due to the different subnets, this action will trigger L3 vrouter function.

# ping ( danny_vm2 )

# ip netns exec qrouter-f1e03fef-cccf-43de-9d35-56d11d636765 tcpdump -eln -i qr-4433f31f-5d icmp

The interface qr-4433f31f-5d is my subnet's gateway port as follows:
# neutron --os-tenant-name danny port-list | grep 4433f31f-5d
| 4433f31f-5d93-4fe4-868a-04ddcc38be20 |                                                 | fa:16:3e:25:22:b3 | {"subnet_id": "d169f180-4304-42f0-b11f-e094287bcd00", "ip_address": ""}  |

Keepalived related

L3HA mode is havily relied on the daemon: Keepalived and this daemon is existed in qrouter namespace.

# vi /var/lib/neutron/ha_confs/f1e03fef-cccf-43de-9d35-56d11d636765/keepalived.conf
vrrp_instance VR_1 {
    state BACKUP
    interface ha-857640ad-a6
    virtual_router_id 1
    priority 50
    garp_master_delay 60
    advert_int 2
    track_interface {
    virtual_ipaddress { dev ha-857640ad-a6
    virtual_ipaddress_excluded { dev qg-f02984c6-dc dev qg-f02984c6-dc dev qr-4433f31f-5d dev qr-16e20a36-fc dev qr-35235c4f-64
        fe80::f816:3eff:fe0d:2702/64 dev qr-16e20a36-fc scope link
        fe80::f816:3eff:fe25:22b3/64 dev qr-4433f31f-5d scope link
        fe80::f816:3eff:fe51:30a1/64 dev qg-f02984c6-dc scope link
        fe80::f816:3eff:fe8f:a85b/64 dev qr-35235c4f-64 scope link
    virtual_routes { via dev qg-f02984c6-dc

There are other two files under /var/lib/neutron/ha_confs/<< qrouter uuid >>/
neutron-keepalived-state-change.log ==> log file
state ==> HA status

# find -L /proc/[1-9]*/task/*/ns/net -samefile /run/netns/qrouter-f1e03fef-cccf-43de-9d35-56d11d636765 | cut -d/ -f5

# ps aux | grep -e "2276895|2276896|2277216|2277217|3284547"
neutron  2276895  0.0  0.0 126160 41364 ?        S    Jul22   0:00 /usr/bin/python2.7 /usr/bin/neutron-keepalived-state-change --router_id=f1e03fef-cccf-43de-9d35-56d11d636765 --namespace=qrouter-f1e03fef-cccf-43de-9d35-56d11d636765 --conf_dir=/var/lib/neutron/ha_confs/f1e03fef-cccf-43de-9d35-56d11d636765 --monitor_interface=ha-857640ad-a6 --monitor_cidr= --pid_file=/var/lib/neutron/external/pids/ --state_path=/var/lib/neutron --user=119 --group=125
root     2276896  0.0  0.0   6696   756 ?        S    Jul22   0:00 ip -o monitor address
root     2277216  0.0  0.0  44752   856 ?        Ss   Jul22   0:13 keepalived -P -f /var/lib/neutron/ha_confs/f1e03fef-cccf-43de-9d35-56d11d636765/keepalived.conf -p /var/lib/neutron/ha_confs/ -r /var/lib/neutron/ha_confs/
root     2277217  0.0  0.0  51148  1712 ?        S    Jul22   0:24 keepalived -P -f /var/lib/neutron/ha_confs/f1e03fef-cccf-43de-9d35-56d11d636765/keepalived.conf -p /var/lib/neutron/ha_confs/ -r /var/lib/neutron/ha_confs/
neutron  3284547  0.0  0.0 172176 36032 ?        S    Jul22   0:00 /usr/bin/python2.7 /usr/bin/neutron-ns-metadata-proxy --pid_file=/var/lib/neutron/external/pids/ --metadata_proxy_socket=/var/lib/neutron/metadata_proxy --router_id=f1e03fef-cccf-43de-9d35-56d11d636765 --state_path=/var/lib/neutron --metadata_port=8775 --metadata_proxy_user=119 --metadata_proxy_group=125 --verbose --log-file=neutron-ns-metadata-proxy-f1e03fef-cccf-43de-9d35-56d11d636765.log --log-dir=/var/log/neutron

# neutron l3-agent-list-hosting-router f1e03fef-cccf-43de-9d35-56d11d636765

Then, we learn that the master vrouter is in node-8. 
There are other ways to know which node is master:
1. use the command to see if the interface qr-xxxxx and qg-xxxxx have ip address or not. If yes, this node is master.
  • ip netns exec qrouter-f1e03fef-cccf-43de-9d35-56d11d636765 ip a
2. Check the following file that contains "master" or not.
  • vim /var/lib/neutron/ha_confs/<< qrouter uuid >>/state
For more details:

Friday, June 24, 2016

[Ceilometer] To survey how to improve the performance of OpenStack Ceilometer

Frankly speaking, OpenStack Ceilometer will suffer some kind of performance issues sooner or later if you don't modify or tune the configuration. The issues has two parts that need you to consider. One is the message bus and API loading, and the other is database. However, I find some best practices which are easy and quick for us to adopt. Here you go:

1. Telemetry(Ceilometer) best practices

a. Data collection

  1. Based on your needs, you can edit the pipeline.yaml configuration file to include a selected number of meters while disregarding the rest.
  2. By default, Telemetry service polls the service APIs every 10 minutes. You can change the polling interval on a per meter basis by editing the pipeline.yaml configuration file.
    for example:
    vim /etc/ceilometer/ceilometer.conf
    => evaluation_interval=120
    vim /etc/ceilometer/pipeline.yaml
    => interval: 120

  3. you can delay or adjust polling requests by enabling the jitter support. This adds a random delay on how the polling agents send requests to the service APIs. To enable jitter, set shuffle_time_before_polling_task in the ceilometer.conf configuration file to an integer greater than 0.

b. Data storage

  1. We recommend that you avoid open-ended queries.
  2. You can install the API behind mod_wsgi, as it provides more settings to tweak, likethreads and processes in case of WSGIDaemon. a. For more information on how to configure mod_wsgi, see the Telemetry Install Documentation.
  3. The collection service provided by the Telemetry project is not intended to be an archival service. Set a Time to Live (TTL) value to expire data and minimize the database size.
    for example:
    vi /etc/ceilometer/ceilometer.conf
    => time_to_live=302400
  4. Use replica sets in MongoDB. Replica sets provide high availability through automatic failover. If your primary node fails, MongoDB will elect a secondary node to replace the primary node, and your cluster will remain functional.
    1. For more information on replica sets, see the MongoDB replica sets docs.
  5. Use sharding in MongoDB. Sharding helps in storing data records across multiple machines and is the MongoDB’s approach to meet the demands of data growth.

2. Metering Service (Ceilometer): Best Practices and Optimization

a. Modifying the List of Meters

    - name: meter_source
      interval: 604800
          - "instance"
          - "image"
          - "image.size"
          - "image.upload"
          - "image.delete"
          - "volume"
          - "volume.size"
          - "snapshot"
          - "snapshot.size"
          - "ip.floating"
          - "network.*"
          - "compute.instance.create.end"
          - "compute.instance.delete.end"
          - "compute.instance.update"
          - "compute.instance.exists"
          - meter_sink
    - name: meter_sink
          - notifier://

b. Modifying the Polling Intervals

The interval attribute is the time between polls. Meters that are available as both notification and polling are going to be polled at the specified interval. To rely on notifications rather than polling, set the interval attribute to 604800 seconds, or once a week.


One of the main issues operators relayed was the polling that Ceilometer was running on Nova to gather instance information. It had a highly negative impact on the Nova API CPU usage, as it retrieves all the information about instances on regular intervals.

Thursday, June 23, 2016

[Linux] Why does Linux require moving IP from eth interface to bridge interface?

This could be a common problem if you have KVM ( or other hypervisor in Linux ) on your physical server and want to use bridge mode with your VMs. At the same time, you also want to let your physical server has the network that can be accessed from other hosts at the same network subnet. At this moment, when a network interface (e.g., eth0) is added to a Linux bridge (e.g., br0), the IP address must be removed from eth0 and added to br0 for the networking to function properly.

I find some answers as follows:

The NIC represents the uplink cable. A cable is layer 1, not layer 3. Now the Bridge works as the device that is being addressed for network traffic (incoming) on the server - either on layer 2 (Ethernet/MAC) and/or layer 3 (IP). So the device that responds to ARP-requests is the bridge - which is good, since it needs to distribute the traffic to the other interfaces on that bridge. If the responding device were the NIC, traffic would not be passed further on to the bridge.

Normally it does not make sense to put any L3 protocol address on port interfaces - because incoming packets are diverted to the bridge interface before the L3 protocol is examined. This means the L3 protocol running on the port interface will never see any incoming packets.

Wednesday, June 15, 2016

[Neutron] Neutron Cheat Cheat Sheet

In recent days, the hand drawing style is becoming more and more pervasive in Taiwan. I don't know how it happens, but at least I can draw the Neutron cheat cheat sheet to echo this kind of style for fun.

P.S: This picture is originally for my colleagues to trouble shooting the Neutron networking problems.