A video alternative to animated GIFs

I wanted to do something like my previous post, but using a video format instead of the ancient .gif format. Here's what I ended up using:

ffmpeg -ss 00:04.20 -i in.mp4 -t 00:06.30 -an -vf "crop=1668:1668:1292:10, scale=480:-1" out.mp4

Options:

• -ss 00:04.20
  Start 4.2 seconds into the input file
• -i in.mp4
  Specifies the input file
• -t 00:06.30
  Limit duration to 6.3 seconds
• -an
  No audio
• -vf " ... "
  Specify the filter chain to operate on the video
• crop=1668:1668:1292:10
  Crop out a 1668 (width) x 1668 (height) window, with the top-left corner located at x=1292, y=10
• scale=480,-1
  Rescale to width=480, height=auto (preserving aspect ratio)
• out.mp4
  Specifies the output file

Making animated gifs

Recently I wanted to make some animated gifs from a video.

First I used my video player (mpc-hc) to grab some frames and save them as .jpg's in a folder.

Then using ImageMagick, I cropped, resized, and created a gif:
$ convert -crop 1200x900+480+0 +repage -resize 480x480 -delay 8 -loop 0 *.jpg output.gif

The repage argument is necessary so that the output gif doesn't use a virtual canvas for the crop operation.

This was kinda big, though, so I looked at this gist and this forum thread to get some ideas on how to reduce the filesize. The source video is live-action footage with a stationary camera, and I settled on:

$ convert -fuzz 5% -ordered-dither o8x8,30 -layers Optimize +map output.gif output2.gif

This dropped the filesize from 1.9 MB to 0.6 MB without affecting the quality too much. The number of colors for posterizing the ordered dither (30) was chosen by some trial and error.

Electric vehicle CO2 emissions

We recently switched to an all-electric vehicle (the Chevy Bolt) and I wanted to investigate how the CO₂ emissions compare to a conventional vehicle.

Choosing colors for scientific visualization: the RGB gamut in CIELUV

I make a lot of plots for work, and color is a very useful way to convey information. These colors are usually represented in RGB, especially when dealing with colors on a computer screen.

However, RGB does not do a good job of representing human perception of color. You will find that [0 255 0] (green) appears much brighter than [255 0 0] (red), which is in turn brighter than [0 0 255] (blue). Furthermore, if you sample uniformly from the RGB space, you will find that an unreasonably large fraction of this appears green.

There have been many attempts to parameterize color in a way that is more closely related to human perception. The one that I've been using is the CIE 1976 (L*, u*, v*) color space, aka CIELUV. The L* is supposed to represent the apparent brightness, and (u*, v*) parameterize the apparent color. It is supposed to do this in such a way that the distance between two points in the CIELUV space corresponds to how distinguishable the corresponding colors are to a human with trichromatic vision.

However, not everything in this color space is representable on a computer monitor. That is, not all of those points will fall within the RGB gamut. So without further ado, I present the RGB gamut in the CIELUV color space:

The boundaries are where one or more of the RGB coordinates saturate.

According to my friend, who (like 7% of American males) has impaired color vision, the color contrast is greatest traveling along the vertical (v*) axis. Of course, there are many ways in which your color vision could be impaired, so his experience is not representative of all color blind people.

More plots below:

Garage door opener repair

A while back, our garage door opener (a Genie Pro 82 screw-drive) developed an odd problem: the light was stuck on. Normally, it is supposed to turn off a few minutes after use, but now it stopped turning off.

I reasoned that it was probably a relay that failed (it's always the moving parts that fail) and that it was stuck in the closed (i.e. "on") position. A quick search online confirmed that this is a common failure mode.

The opened opener:

Printing on both sides

This keeps annoying me; I don't know why it keeps coming up.

My printer has a duplexer, but sometimes Windows 7 forgets about it. HP has a help page about it, the gist of which is:

Control Panel -> Right-click on your printer -> Printer properties -> Device Settings

Set "Duplex Unit" to "Installed" and, to reduce confusion, set "Allow manual duplexing" to "Disabled"

Terminal adventures

I do a lot of work remotely for work. Somewhere in the server room there is a beast of a computer with 100's of GB of RAM and a 10-gigabit connection to a dozen other computers hosting half a petabyte of storage.

Mostly I work in MATLAB. If I need graphics, we have a nice OpenNX setup that does remote desktop quite well. It's nearly seamless when I'm on-site, and is still tolerable when I use it from home over DSL (as long as nobody else at home is using the upload capacity).

But often I don't need full graphics, just a text editor and the MATLAB command window. So in today's adventure, we are going to try setting up tmux and vim.

Xbox 360: 120mm fan mod for quietness

I wanted to watch a movie yesterday, but my Xbox's fans are too loud. So I decided to do something about it!

Batch photo resize preserving modify date

On Mac OS with ImageMagick installed (see also this older post):

mkdir resized
cp 2013-09-02*.jpg resized
mogrify -resize 3056x3056 resized/*.jpg
find resized -name "*.jpg" | sed "s/resized\///" | xargs -I {} touch -r "{}" "resized/{}"
mv resized/*.jpg .
rmdir resized

The crazy find/sed/xargs thing does the following:

1. find : Find all the jpg files in the "resized" folder and return their relative path.
2. sed : Find and replace all instances of "resized/" with an empty string. sed returns the full string with replacements, not just the match.
3. xargs touch : Change each of the files in the "resized" folder to have the same modify date

Git rebase

After seing git rebase show up on StackOverflow a few times and purposely ignoring it (due to the overhead required to learn a new git command), I finally broke down and learned what it does.

And it turns out it's not that bad! The git book has a good page introducing it with nice pictures:

http://git-scm.com/book/en/Git-Branching-Rebasing

and I'll describe how I applied it to my particular situation.

Scenario

I have been pulling from a repository at origin/develop, but also making local changes and committing them to my local repository. Now I want to push some (but not all) of the changes to origin/develop.

Procedure

First, I wanted to see how my local repository (currently on branch develop) differed from origin/develop:

git diff --name-status origin/develop..develop

Now if I wanted to make all of those changes, I can just do git push origin, and it would be done. Instead I want to make a subset of those changes, but I can't just arbitrarily pick changes because maybe some of them depend on previous commits.

That's how rebase helps: it lets us re-order the commits. It puts all of the remote changes first (this makes pushing trivial), and then it applies our local changes in whatever order we tell it to. So we do:

git rebase -i origin/develop

and this opens up a text editor with all the local changes that have happened since we first diverged from origin/develop. We can rearrange the commits, collapse multiple commits into one using squash, and do all sorts of other things that are covered in this tutorial. In my case, I simply moved the changes I wanted pushed to the top of the list.

Once we save and close this file, git-rebase will try to apply each of those changes, in sequence, using origin/develop as a starting point. This may not proceed entirely smoothly, especially if you have re-ordered some commits.

Once this is finished, I go find the hash of the last commit I want pushed and push all the commits up to that one:

git push origin 1fc6c95:develop

And that's that!