Convert ppm files to png
mkdir png; for i in *.ppm; do j=${i/ppm/png}; echo $j; convert $i -depth 8 $j; mv $j ./png/; done

for j in *.ppm; do cd $j; mkdir png; for i in *.ppm; do j=${i/ppm/png}; echo $j; convert $i -depth 8 $j; mv $j ./png/; done; cd ..; done

Convert to Grayscale Image
for i in *.png; do convert $i -type Grayscale -depth 8 $i; done

Resize the png image from 640×480 to 320×240
for i in *.png; do echo $i; convert $i -filter Cubic -resize 320×240 -depth 8 $i; done

// the point filter is supposed to only use colors from the original image to produce the smaller one, however, this doesn’t seem to work all the time
for i in *.png; do echo $i; convert $i -filter point -resize 320×240 -depth 8 ../labels_320x240/$i; done
// use the “sample” function instead:
for i in *.png; do convert $i -sample 50% -depth 8 ../$i; done

Resize all images to a max size (along any dimension) to be 320
mogrify -resize 320×320 -depth 8 *.png

Rename files sequentially from 00001 to 99999
let x=0; for i in *.png; do echo $i; let “x += 1″; echo $x; mv $i $(printf “%05d.png” ${x}); done

Rename and make video for previewing
name=”prefix_”; let x=0; for i in *.ppm; do echo $i; let “x += 1″; echo $x; mv $i $name$(printf “%05d.ppm” ${x}); done; ffmpeg -r 15 -i $name%05d.ppm -b 800k -g 300 -an ${name%%_}.mp4

Convert 16-bit files to 8-bit.
for i in *.png; do echo $i; convert $i -depth 8 $i; done

Make Apple Quicktime compatible MP4 file for viewing
(a) ffmpeg -r 15 -i %05d.ppm -an video.mp4
(b) ffmpeg -r 6 -i bus_cif%06d_blended.png -an -r 24 video.mp4

Renaming and moving all files
let a=5; for i in *.png; do j=”../../labels/prefix${a}_”; k=”$j$i”; echo $k; mv $i $k; done

Renaming and moving every 50 files
let a=3; let x=1; for i in *png; do let “j = $x % 50″; if [ $j == 1 ]; then k=”../../frames/prefix${a}_${i}”; mv $i $k; fi; let “x += 1″; done

Retaining every 1 of N frames
let n=0; ext=”.png”; for label in *_label.png; do ID=${label%%_label.png}; name=$ID$ext; let “mod = $n % 2″; if [ $mod != 0 ]; then rm $label; rm $name; fi; let “n += 1″; done
Alpha Blend a single image
composite -blend 50 -gravity South img1.png img2.png

Alpha blend all images in folder
(a) for labelingResult in *-rgb.png; do inputImage=${folder/}; outputImage=${inputImage%%.png}_blend.png; composite -blend 50 -gravity South $labelingResult $inputImage $outputImage; done;

(b) mkdir blend; for i in *GT.png; do composite -blend 50 -gravity South -depth 8 $i ${i/_GT.png/.png} ./blend/${i/_GT.png/_blended.png}; done;

Scale images 200% so that its easier to see pixel-wise labels
for i in *.png; do convert $i -scale 200% -depth 8 $i; done

Convert from RGBA image to RGB image
for i in *.png; do convert $i -flatten +matte -depth 8 $i; done

Make montage for video
for i in *_co.png; do montage -background transparent -geometry +4+4 -depth 8 $i “../folder/”${i/_co.png/_vgs.png} “../mjrd5_montage/”${i/_co.png/_montage.png}; done

Make videos using “ffmpeg” from a sequence of frames
mkdir temp; let x=1; for i in propagated_blended_*.png; do echo $i; y=$(printf “./temp/%05d.png” $x); echo $y; cp $i $y; let “x += 1″; done; ffmpeg -r 3 -i ./temp/%05d.png -vcodec libx264 -crf 25 -threads 0 -an video-h264.avi; cp video-h264.avi ~/Desktop/videos; mv video-h264.avi ..; rm -r temp

let x=1; for i in *.png; do mv $i $(printf “%05d.png” ${x}); let “x += 1″; done

Run MATLAB in batch and in background from command line
nohup /vpml/bin-osx/MATLAB_R2010a.app/bin/matlab -nojvm -nodisplay < /vpml/scratch/aychen/label_propagation/work/runall.m >& /vpml/scratch/aychen/label_propagation/work/runall.log &

Duplicate an image (label) N times and increase its index sequentially
x=0; y=1; while [ $x -le 84 ]; do cp $(printf “propagated_%06d.png” ${x}) $(printf “propagated_%06d.png” ${y}); let “x += 1″; let “y += 1″; done

Blend the duplicated images (labels) with the input images
x=0; while [ $x -le 84 ]; do composite -blend 50 -gravity South $(printf “propagated_%06d.png” ${x}) $(printf “folder/prefix_cif%06d_gray.png” ${x}) $(printf “propagated_blended_%06d.png” ${x}); let “x += 1″; done

Histogram Equalization on a single image (for display purposes)
convert $i -equalize $i

Modify Brightness/Saturation/Hue of an image (e.g. +30% brightness, -10% saturation, don’t change Hue)
convert $i -modulate 130,90. $i

Well, because I need to work secretly on research ideas before they get published, and because computer science is a very practical field, and I believe a successful and productive researcher also needs to be a skilled programmer. Also, since I’m currently in charge of overseeing our in-lab code development and so many lab-members are contributing to our in-house library everyday, I feel like going through these issues before the library gets way too messy to be cleaned up.

Coding Styles

A good blog discussing about C++ coding style:

http://hulubei.net/tudor/articles/Coding-Style-Abstract.html

Although not all of its coding style suggestions are agreed by my adviser and I (e.g. using the Hungarian notation while naming variables, using the BSD/Allman indent style other than the K&R indent style). For a complete list of indent styles, please refer to the following link:

http://en.wikipedia.org/wiki/Indent_style

Hard/soft tab conversion in Eclipse

Although one can set Eclipse to put spaces for tabs while editing, it doesn’t have a built-in function for converting existing files from tabs (hard tabs) to spaces (soft tabs). Please use the award winning Eclipse plugin, “AnyEdit”

http://andrei.gmxhome.de/anyedit/

Having a base_class B and an derived_class D, where D has a function that returns a more specific (derived) class than the base class B.

For example, base_class has a clone function

base_class* base_class::clone()

that returns a pointer to base_class B. If we want to use clone() in D to obtain a pointer to the newly created derived_class D’, the function D.clone() will have covariant return type from B.clone().

Since covariant return type was added to the C++ standard some time later in the its development, older C++ compilers (VC++6 and older, GCC 3.4 and older) do not support covariant return types. The only way to implement clone() in derived_class without using covariant return type is to upcast D’ to base_class before returning it, and later on, if we want to use derived_class specific functions, we will have to downcast* the pointer back, which is slower and troublesome.

With covariant return type, a derived class can return a data type that is more derived/specific than the one it overrides. In the previous example, we will be able to make derived_class::clone() return derived_class* while B.clone() still return base_class*. However, there are restrictions to covariant return types in C++:

• The function D::f returns a reference or pointer to a class of type T, and A::f returns a pointer or a reference to an unambiguous direct or indirect base class of T.
• The const or volatile qualification of the pointer or reference returned by B::f has the same or less const or volatile qualification of the pointer or reference returned by A::f.
• The return type of B::f must be complete at the point of declaration of B::f, or it can be of type B.

Notes:

* Compilers usually won’t allow static downcast in the code, therefore we’ll have to use dynamic cast which is done during runtime. Please refer to my older post for dynamic casting.

References:

http://publib.boulder.ibm.com/infocenter/comphelp/v8v101/index.jsp?topic=/com.ibm.xlcpp8a.doc/language/ref/cplr139.htm

http://www.parashift.com/c++-faq-lite/virtual-functions.html#faq-20.8

http://www.lwithers.me.uk/articles/covariant.html

Some might encounter “invalid project path: include path not found” errors after the move, even if all the include paths are correct in the hand-tailored makefile. Its likely caused by the “automated discovery of paths and symbols” option, located at the project’s “propeties -> C/C++ build -> Discovery Option”. The path to the old directory location is still being included during compilation, thus causing the error. A typical solution is to stop using the automated discovery of paths and symbols function and remove the existing paths

http://www.saltycrane.com/blog/2007/02/how-to-remove-discovered-include-paths/ .

However, for those that are happy with the automated function and simply want to fix the paths for one time, just go to the hidden folders in the current workspace directory

workspace/.metadata/.plugins/org.eclipse.cdt.make.core

and delete the “projectname”.sc file. When Eclipse is started next time, it will re-detect the current paths and symbols—which would not contain the old paths—and thus the problem is solved.

http://cvschangelog.sourceforge.net/

(from http://www.smalltalk.org/alankay.html)

OS X is a very nice operating system indeed, with both the stability and safety of the freebsd core it was built on, and the user-friendly Macintosh GUI. Most linux/unix open source software can run directly on it without a problem, but the current build of Transcode is an exception. Transcode, a pretty powerful image/audio/video converting tool, makes use of a wide combination of libraries, such as ffmpeg, imagemagick, … etc. This makes it capable of dealing with almost any file format, but this also brings a lot of dependency problem to it. Transcode needs ffmpeg, ffmpeg requires gtk2, yet the compilation of gtk2 on OS X would require a later version of X11. Unfortunately, the last time Apple updated their X11 was in 2006, so we’d have to patch this problem manually with the unofficial release of X11 in order to install all these packages. One would still have to do this even when OS X package management tools such as Fink is used.

24 commonly used sequences, such as Akiyo, Carphone, Coastgaurd, Foreman, Hallway, Salesman, Silent, Stephan, and Suzie. In both QCIF (176×144) and CIF format (352×288).

Sequences Collected by Xiph.org

43 sequences. Including most of the above and some more, such as Flower (Garden), Students, Tennis. 11 of them are HD sequences (720p or 1080p).