| gra |
| |
- Classification:Numerical Algorithm-Artificial Intelligence - matlab
- Development Tool:matlab
- Sise:84.0 KB
- Upload time:2011/6/16 3:20:29
- Uploader:naka
- Download Statistics:
|
| Description |
image processing via webcam  File list:
as.jpg
find.m
If you are a member, Log in. If you are not a member, Please register
|
| Related source code |
[BBO Biogeography-based optimization ...] - Biogeography-based optimization (BBO) software for minimizing a general function
[ch 4 javahow to pogram] - this code of java how to program
[fruit grading] - This code is for Fruit Grading. It takes the Fruit image and delect its damage parts and then grade it
[rgb2gray sysgen] - rgb2gray algotrith xilinx system generator
[RSA program] - RSA program
[> compile
Remember to also adjust your path so MATLAB can find iHOG:
>> addpath(genpath('/path/to/ihog'))
If you want to use iHOG in your own project, you can simply drop the iHOG directory into the root of your project.
Inverting HOG
To invert a HOG point, use the 'invertHOG()' function:
>> feat = features(im, 8);
>> ihog = invertHOG(feat);
>> imagesc(ihog); axis image;
Computing the inverse should take no longer than a second for a typical sized image on a modern computer. (It may slower the first time you invoke it as it caches the paired dictionary from disk.)
Visualizing HOG
iHOG has several functions to visualize HOG. The most basic is 'visualizeHOG()':
>> feat = features(im, 8);
>> visualizeHOG(feat);
The above displays a figure with the HOG glyph and the HOG inverse. This visualization is a drop-in replacement for more standard visualizations, and should work with existing code bases.
The de-facto HOG has signed components, unsigned components, as well as texture components. 'dissectHOG()' visualizes each of these components invidually:
>> dissectHOG(feat);
A similar visualization 'spreadHOG()' shows each dimension individually:
>> spreadHOG(feat);
More visualizations are available. Check out the 'visualizations/' folder and read the comments for more.
Learning
We provide a prelearned dictionary in 'pd.mat', but you can learn your own if you wish. Simply call the 'learnpairdict()' function and pass it a directory of images:
>> pd = learnpairdict('/path/to/images/', 1000000, 1000, 5, 5);
The above learns a 5x5 HOG patch paired dictionary with 1000 elements and a training set size of one million window patches. Depending on the size of the problem, it may take minutes or hours to complete.
Bundled Libraries
The iHOG package contains source code from the SPAMS sparse coding toolbox (http://spams-devel.gforge.inria.fr/). We have modified their code to better support 64 bit machines.
In addition, we have included a select few files from the discriminatively trained deformable parts model (http://people.cs.uchicago.edu/~rbg/latent/). We use their HOG computation code and glyph visualization code.
Questions and Comments
If you have any feedback, please write to Carl Vondrick at vondrick@mit.edu.
References
The conference paper for this software is currently under submission. In the mean time, please see our technical report:
[1] Carl Vondrick, Aditya Khosla, Tomasz Malisiewicz, Antonio Torralba. "Inverting and Visualizing Features for Object Detection." Technical Report. 2013."" target="_blank">iHOG: Inverting Histograms of Orient...] - iHOG: Inverting Histograms of Oriented Gradients
This software package contains tools to invert and visualize HOG features. It implements the Paired Dictionary Learning algorithm described in our paper "Inverting and Visualizing Features for Object Detection" [1].
Installation
Before you can use this tool, you must compile iHOG. Execute the 'compile' script in MATLAB to compile the HOG feature extraction code and sparse coding SPAMS toolbox:
$ cd /path/to/ihog
$ matlab
>> compile
Remember to also adjust your path so MATLAB can find iHOG:
>> addpath(genpath('/path/to/ihog'))
If you want to use iHOG in your own project, you can simply drop the iHOG directory into the root of your project.
Inverting HOG
To invert a HOG point, use the 'invertHOG()' function:
>> feat = features(im, 8);
>> ihog = invertHOG(feat);
>> imagesc(ihog); axis image;
Computing the inverse should take no longer than a second for a typical sized image on a modern computer. (It may slower the first time you invoke it as it caches the paired dictionary from disk.)
Visualizing HOG
iHOG has several functions to visualize HOG. The most basic is 'visualizeHOG()':
>> feat = features(im, 8);
>> visualizeHOG(feat);
The above displays a figure with the HOG glyph and the HOG inverse. This visualization is a drop-in replacement for more standard visualizations, and should work with existing code bases.
The de-facto HOG has signed components, unsigned components, as well as texture components. 'dissectHOG()' visualizes each of these components invidually:
>> dissectHOG(feat);
A similar visualization 'spreadHOG()' shows each dimension individually:
>> spreadHOG(feat);
More visualizations are available. Check out the 'visualizations/' folder and read the comments for more.
Learning
We provide a prelearned dictionary in 'pd.mat', but you can learn your own if you wish. Simply call the 'learnpairdict()' function and pass it a directory of images:
>> pd = learnpairdict('/path/to/images/', 1000000, 1000, 5, 5);
The above learns a 5x5 HOG patch paired dictionary with 1000 elements and a training set size of one million window patches. Depending on the size of the problem, it may take minutes or hours to complete.
Bundled Libraries
The iHOG package contains source code from the SPAMS sparse coding toolbox (http://spams-devel.gforge.inria.fr/). We have modified their code to better support 64 bit machines.
In addition, we have included a select few files from the discriminatively trained deformable parts model (http://people.cs.uchicago.edu/~rbg/latent/). We use their HOG computation code and glyph visualization code.
Questions and Comments
If you have any feedback, please write to Carl Vondrick at vondrick@mit.edu.
References
The conference paper for this software is currently under submission. In the mean time, please see our technical report:
[1] Carl Vondrick, Aditya Khosla, Tomasz Malisiewicz, Antonio Torralba. "Inverting and Visualizing Features for Object Detection." Technical Report. 2013.
[Fortran&C programs for the time-depe...] - Here we present numerical algorithms for both stationary and non-stationary solutions of the full threedimensional (3D) GP equation for a dipolar BEC, including the contact interaction. We also consider the simplified one- (1D) and two-dimensional (2D) GP equations satisfied by cigar- and disk-shaped dipolar
BECs. We employ the split-step Crank–Nicolson method with real- and imaginary-time propagations, respectively, for the numerical solution of the GP equation for dynamic and static properties of a dipolar
BEC.
[Approximating integral by (gauss qua...] - Approximating integral by (Gauss quadrature)or (newton cotes)
[batch file programming to check pass...] - this program used to verify the predefined password. it can be used to define password to any folder
[Atmospheric and Gravity Model for Ai...] - Included are two Matlab functions: atmos_model and gravitywgs84model.
Atmos_model calculates the gas properties in the 1976 Standard Atmosphere. Given a height the function returns density, speed of sound, temperature and kinematic viscosity.
Gravitywgs84model computes the gravity constant at a specific geodetic latitude and height using the WGS 84 model.
|
| Download Address |
 DownLoad
|
| Comments: Don't forget to comment after downloading! Comment... |
|
