Desert Rose: My Research

Showing posts with label My Research. Show all posts

Oct 21, 2017

Perspective transformation nitty gritty

Feb 8, 2016

Indoor scene dataset collection

Berkeley B3DO

849 images.
50 categories.
Bounding box level annotation.

SUN 3D

polygonal 2D semantic annotation of total of 10335 RGBD images. (additional 3D polygon for 3D detection is also available, 3D polygonal for room layout annotation.)
4943 new images captured.

3784 images using (Kinect v2).
1159 using (Intel Realsense ).

1449 from NYUD v2 (Kinect v1).
554 from B3DO (Kinect v1).
3389 selected frames from SUN3D videos (Asus Xtion).
19 scene-types (includes most of the NYUD v2 scenes).
Reference: semantic segmentation different CNN models

Robot in a room

Not available.

NYUD v1

7 scenes-types with full labels.
12 object classes to classify.
total 2347 images.

NYUD v2

9 dominant scenes-types.
37 dominant categories of object classes.
total 1449 images.

Cornell RGBD

Office and home scene-types
28 office scenes (each scene has point cloud annotated bag file)
24 home scenes
separate image with label for each frame is available.

COLD (Ljubljana+Freiburg+Strucken)

5 common scenes (bathroom, kitchen, printer, office, hallway) in all three geo-locations. Additionally other scenes as sequence images.
RGB only

karpathy Discovery

58 scenes (like Cornell RGBD) of office+kitchen+printer

36 office desks + 7 bookshelves + 4 printers + 3 kitchen coutners + 8 miscellaneous living space scenes.

Types of objects are different than that of NYUD v2; more object centric.
3D meshes representations of the scene from kinect fusion + post-processing.

paspart_challenge: part segmentation

Oct 21, 2015

Inference algorithms in Graphical Model

Inference algorithms in undirected graphical model or markov random field categories:

Propagation based

Loopy belief propagation

Variational method based (approximate not exact)

Mean field
Advantage

Intuition is that complex graphs can be probabilistically simple; in particular, in graphs with dense connectivity there are averaging phenomena that can come into play, rendering nodes relatively insensitive to particular settings of values of their neighbors. This average phenomena leads to simple inference algorithm.

Monte Carlo method based

Gibbs sampling based
Advantage

simple to implement
theoretical guarantee to converge

Disadvantage

slow to converge
hard to diagnose

Inference algorithms:

Exact inference by the Junction-Tree algorithm (Lauritzen and Spiegelhalter, 1988)
Loopy Belief Propagation (Pearl, 1988)
Generalized Belief Propagation (Yedidia et al., 2005)
Tree Re-weighted Belief Propagation (Wainwright et al., 2005)
Propagation based on convexification of the Bethe free energy (Meshiet al., 2009).
Mean field (Jordan et al., 1998)
Gibbs sampling (Geman and Geman, 1984)

Sep 22, 2015

NIPS papers 2015

Acquiring Visual Classifiers from Human Imagination. Carl Vondric et al.
3D Object Proposals for Accurate Object Class Detection. Utrasun et al.
Precision-Recall-Gain Curves: PR Analysis Done Right. Milis Kull et al.
Unsupervised Learning by Program Synthesis. Josh Tenenbaum et al.
Deep Visual Analogy-Making. Honglak Lee et al.
Lifelong Learning with Non-i.i.d. Tasks. Christopher Lampert et al.
*Semantic Component Analysis (ICCV-15)

Mar 13, 2015

Oct 10, 2014

Deep Learning: Caffe installation ubuntu 12.04

This is a concise, concrete, and step-wise instruction for installing caffe in ubuntu 12.04 (guided by the instructions given the caffe authors).

Followings are the prerequisite libraries. Carefully install it all.

atlas:

sudo apt-get install libatlas-base-dev
installs into "/usr/include/atlast"

boost:

sudo apt-get install libboost-all-dev
installs into "/usr/include/boost"

OpenCV:

sudo apt-get install libopencv-dev

hd5:

sudo apt-get install libhdf5-serial-dev
install into "/usr/lib/"

Addtionally, if python pip is not installed install it using aptitude.

anaconda:

Download and follow instruction.
Download link
bash Anaconda-2.1.0-Linux-x86_64.sh

mdb:

download the mdb from github
cd mdb-mdb/libraries/liblmdb
make
sudo make install

google-log:

wget https://google-glog.googlecode.com/files/glog-0.3.3.tar.gz
cd glog-0.3.3
./configure
make
sudo make install

gflags:

wget https://github.com/schuhschuh/gflags/archive/master.zip
cd gflags-master/
mkdir build && cd build
export CXXFLAGS="-fPIC" && cmake .. && make VERBOSE=1
make
sudo make install

numpy:

aptitude search numpy
sudo aptitude install python-numpy
install into " /usr/include/numpy"

All the pre-requisites are installed, now we can proceed to install the caffe.

download the caffe
go into the caffe-master
copy and rename the Makefile.config
do the following change into the Makefile.config

# CPU-only switch (uncomment to build without GPU support).
CPU_ONLY := 1
# MATLAB directory should contain the mex binary in /bin. MATLAB_DIR := /usr/local
PYTHON_INCLUDE := /usr/include/python2.7 \
/usr/lib/python2.7/dist-packages/numpy/core/include

make all
make test
make runtest

Try out the pre-trained ImageNet model to test the caffe using python in the following reference.

Expected issues:

append the PYTHONPATH into your ~/.bashrc file with. The following two instruction will do it for you.

make python
make distribute

anaconda GLIB issue:

Error: ImportError: `GLIBC_2.15' not found caffe
Solution: delete the 'libm.so and libm.so.6' from ~/anaconda/lib/

protobuf issue:

Error: caffe ImportError: No module named google.protobuf caffe
Solution:

find the path where the google-protobuf is installed. for example if it install into "/usr/lib/python2.7/dist-packages/" then do the following into your script.
import sys
sys.path.append("/usr/lib/python2.7/dist-packages/")

Oct 1, 2014

git

    open a git account
    ------------------------

   get git and set your system up with git:
---------------------------------------------------

install using apt-get:

get the requisites:

sudo apt-get install libcurl4-gnutls-dev libexpat1-dev gettext libz-dev libssl-dev build-essential

get git:

wget https://git-core.googlecode.com/files/git-1.8.1.2.tar.gz

unzip :
```
tar -zxf git-1.8.1.2.tar.gz
```
```
 
```
go into the directory for installation
```
cd git-1.8.1.2 
```
```
 
```

install :

make prefix=/usr/local all
sudo make prefix=/usr/local install

setup your github account into the git installed in your system

git config --global user.name "alimurreza123"
git config --global user.email alimurreza123@yahoo.com

If everything works fine you will be able to see the already set user info using the following

git config --list

It should show:

user.name=alimurreza123
user.email=alimurreza123@yahoo.com

Reference

Create a sample project to push into the repository:
-------------------------------------------------------------------

create a workspace in your machine and get into the base directory of the project named 'testing':

mkdir -p ~/git/testing

cd ~/git/testing

copy few files and directory there. In real situation you will have your project directories such data, output, src etc.
converting the existing project into workspace environment

git init

It should show the following message:

Initialized empty Git repository in /home/user/git/testing/.git/

add everything inside the 'testing' project

```
git add .
```

Push the files using commit

git commit -m "Initial Commit" -a

It says you have committed all the files/folders (-a option) and your message is "Initial commint" (-m) option

Alternately you can add files folders explicitly using by their names

If you want to push it into some remote server you need to push the project using 'push' from git.

git remote add origin ssh://git@github.com:alimurreza123/testing.git

If the above does not work try the below:
git remote set-url origin https://github.com/alimurreza/testing

Now push the project 'testing' into github repository

git push origin master

It should show the something similar like this:

Counting objects: 6, done.
Delta compression using up to 8 threads.
Compressing objects: 100% (6/6), done.
Writing objects: 100% (6/6), 2.06 KiB, done.
Total 6 (delta 0), reused 0 (delta 0)
To git@github.com:alimurreza123/testing.git
 * [new branch]      master -> master

If the above does not work try force push as follows:
git push -f origin master

"git push tells git that we want to push our changes, "origin" is the name of our newly-configured remote server and "master" is the name of the first branch."

You may need to setup you ssh-key before you can do this. Following talks about it.
Reference:
Add a folder

git add matlab_src
git add matlab_src/*
git commit
git push

Remove files

git rm afile.m
git commit -m "removing my file"
git push

Update files

git pull

set url

git remote set-url origin https://github.com/alimurreza/electro_migration_failure_pred

git commit -m "Initial commit"
git push
ssh-keygen -t ed25519 -C alimurreza@gmail.com
pbcopy < ~/.ssh/id_ed25519.pub
Go to github-->settings-->SSH & GPG Key

then just paste it

Add the ssh-key if you want to push your code into the repository
using ssh:
-------------------------------------------------------------------------------------

generate the public-private key pair
Add the public key in your github settings
select the paraphrase wisely. You will be prompted to enter the paraphrase when you want to push files into remote repository such as github.
cloning a repository

git clone https://github.iu.edu/mdreza/electro_migration_failure_pred.git

for cloning into iu server:

git clone https://github.com/alimurreza/electro_migration_failure_pred.git
git checkout master

Reference

Other reference:

May 21, 2014

Latex stuff

Use hyphen in math mode

$Non\mbox{-}object$

Multiline equation: \usepackage{amsmath}

\begin{equation}
\begin{split}
... \\
\end{split}
\end{equation}

displaying-a-wide-figure-in-a 2 column doc

trick is to use \begin{figure*} instead of \begin{figure}

Apr 22, 2014

Creating a good presentation

www.principiae.be

Feb 23, 2014

SVM: helps for libsvm

full-example-of-multiple-class-svm-with-cross-validation-using-matlab

Dec 17, 2013

Useful information from Prof. Jeff Offut

Nov 2, 2013

Drone programming

I spent a little time learning how to program a Quadcoptor or Hexacoptor UAVs. We had a small workshop today at GMU. I just started learning how to fly/move/land etc a quadcoptor UAV (Parrot AR Drone 2.0). Quick summary to get things started:

Install node.js (build on javacript runtime to make network application)

Proceed to installing it with the following steps in terminal
./configure
make
sudo make install

Download ar-drone library (An implementation of the networking protocols used by the Parrot AR Drone 2.0.)

You can do it by another software npm (Node Packaged Modules) which allows user to get node.ja application via command line. Install it first (if you don't have it), then get the ar-drone library using npm:
sudo apt-get install npm
npm install ar-drone

Now create a simple program e.g., test_drone.js which will allow the ar-drone to takeoff and stay hovering for some seconds then land down on earth.

var arDrone = require('ar-drone');
var client = arDrone.createClient();
client.takeoff();

client.after(8000, function() {
    this.stop();
    this.land();
   });

Run the javascript from command prompt

node test_drone.js

Congratulations your drone is flying. You can continue building (see the tutorial) on the program to do some more fun stuffs such as receiving the video output from the drone and make an amazing video of your campus. Here (taken from Quadcoptor and Hexacoptor UAVs) is one that is done in our George Mason University, Fairfax campus.
Other reference

Nov 1, 2013

Visualizing octomap in rviz

plotting-a-markerarray-of-spheres-with-rviz

they suggested to get the electric version or later version of rviz where there is MarkerArray type display.
octomap_server publishes topic occupied_cells_vis_array. I was unable to visualize it. In rviz (diamond back version), there is no display type called MarkerArray. There is Markers type display, that I can add from the display list. Finally, I added Markers type display with topic name occupied_cells_vis (notice I omitted _array as suggested by others) and I can visualize the octomap in rviz.

Add display item with MarkerArray with topic occupied_cells_vis (rviz snapshot)
rviz tutorial
octomap-server-rviz-mapping:

octomap_server subscribes to topic cloud_in.

octomap_server related questions and answers
octomap-not-updating-the-map-in-real-time
marker-array-not-visible-in-rviz-but-is-getting-published

Try rostopic echo normals_marker_array to see if messages are actually being published.
I did something similar: rostopic echo occupied_cells_vis_array to see the message in the topic occupied_cells_vis_array published by octomap_server

octomap_server

Octomap linking issue resolved

OctomapGenerator node was not building under ROS. There was linking issue
with the octomap package used by the node.
export octomap_DIR=~/octomap-distribution/lib/cmake/octomap
rosmake rgbd_mapping OctomapGenerator

Oct 31, 2013

Cmake workflow

This process is run for each CMake project.

# In a CMake project
$ mkdir build
$ cd build
$ cmake ..
$ make
$ make install  # (optionally)

CMakeLists.txt

Oct 30, 2013

Sell your idea

how-to-sell-ideas-like-gladwell
Three factors works for his way:
1. KISS (Keep it simple and stupid).
2. Story telling.
3. Introduce curiosity gap; don't just give the answer right away.

Desert Rose