Compression Testing Mobile Video Production Methods

The focus of the Pocketcine project has been short videos that can be downloaded quickly and inexpensively over the air to cell phones.

The smaller the file size, the fewer bytes would flow over the air, triggering smaller carrier charges. Impatient mobile consumers will not be left tapping their toes, waiting for big video files to land on the storage media of their cell phones before playing. One of the main reasons why the world’s most successful viral mobile video, The Crazy Frog, is so widely distributed is that it is only about half a megabyte in size.

From the beginning my research told me that 3D animated shorts would compress most efficiently and live action video the least efficiently. Flash generated vector art and traditional hand-drawn animation would fall into the middle of the spectrum. In an upcoming blog, I’ll publish a primer on MPEG-4 compression so that you will understand the reasons why. However some of the main reasons are that a 3D animation does not have the sharp edges that compression algorithms don’t compress well, it has smoother and less random gradations, its color space works well with the compression algorithms, and the artist has more control over backgrounds and movement than possible with live action video.

I commissioned four videos to be created. The videos had to be short: less than three minutes. And each of them was produced using different creation tools. Take a look at them on YouTube. They have been optimized for broadband playback: (Please leave a comment on the videos!)

1. 3D Animation: Floyd the Fruitfly, a 3D animated mobile ad for Genome BC is 33 secs long. It was entirely generated in 3D software and edited in Adobe Premiere, a video editing program. It has voice and sound effects.

http://www.youtube.com/watch?v=k-tssHcrFu0

2. Flash Vector Art: Evil Rainbow, a music video, was created in Flash, and featured original music. The Flash animation was exported as a video and edited in Premiere.

http://www.youtube.com/watch?v=CqVKFACKYrE

3. Traditional 2D Art: Damsel in Distress is a hand-drawn animated short, edited in Premiere.

http://www.youtube.com/watch?v=WiXKLGd0-sA

4. Live Action Video: Worst Case Dating was shot with a high end consumer video camera and edited in Premiere.

http://www.youtube.com/watch?v=BmVWvYjYPtg

Floyd was carefully designed to compress well. The background is homogenous and blurry, there is a single character and the soundtrack has little music or sound complexity. It is the measuring stick for the other videos.

The running time of the videos varied from 33 seconds (Floyd) to 2 minutes and twenty seconds in the case of Evil Rainbow. The Floyd Fruit Fly video compressed from an original size of 800 megabytes to an .mp4 that was less than ½ megabyte, a compression ratio of 1700 to 1. And it looked fabulous on the cell phone at 320x240 pixel resolution. The longest video, Evil Rainbow (141 seconds, 4 and a half times as long in running time) and 5 and half times the file size.

I used Sorenson Media’s encoder, Squeeze, to compress each of the four videos. In each case I compressed the raw, uncompressed Adobe Premiere file to a screen resolution of 320x240. In each case I used the exact same compression parameters for the video and the sound track. The only setting I changed was the target size of the output file. Using the Floyd video as the benchmark, I simply increased the output file size according to the length of the video. If the Evil Rainbow video at 141 seconds represents 100%, then Floyd is 23% of the running time, The Damsel in Distress 62%, and the Worst Case Dating video 65%. This means each of the videos got the same amount of compression per unit of time. However, I should note that each video did get optimal compression for its type, since the type of compression was two pass, variable bit rate. Basically I was telling the compression software, “Here is your file size budget, do the best you can to create a compressed file within that limit.”

Video and still compression relies on the fact that it can take out of the source file information that humans can’t see or hear well, if at all. So the fact that the results of the test are judged by the eye rather than with a scientific instrument does not invalidate the test. Here is what we got when we compressed the above four videos with the compression settings.

www.pocketcine.com/muse/Floyd_test.mp4
www.pocketcine.com/muse/Evil_test.mp4
www.pocketcine.com/muse/Damsel_test.mp4
www.pocketcine.com/muse/Worst_test.mp4

Compare them to the YouTube versions.

Lessons Learned
The Floyd video looks the best, but the surprise was that at the same rate of compression, the 2D animation (Damsel in Distress) also compressed really well. Unfortunately for 2D artists, 3D animation is the better long term production solution, especially for videos featuring strong and engaging characters, because the objects you build in 3D software are reusable, whereas 2D animation frames rarely are reused. You simply re-animate a 3D character, place it in a new set or move the camera about.

The Evil Rainbow music video did not compress as well as I expected. Initially I thought this was because it has a very dynamic and continuous soundtrack, eating up a substantial part of the file budget. However, when I compressed it without sound:

www.pocketcine.com/muse/Worst_nosound_test.mp4

I did not get a big improvement in visual quality. Obviously the hard lines of vector art do not compress well. So the conversion of Flash art to digital video does not look like a good strategy for producing mobile videos with very small sizes. (It was not designed for that, but rather to be played back as vector art in the Flash player.) The advantage of Flash production is that it combines some of the advantages of 3D animation (reusable objects) and some of the advantages of 2D animation (more flexible). The problem was that the Evil Rainbow’s artist used a lot of animation to tell his story, which was hard to compress.

The live action video compressed the worst. This is to be expected. Live action video has a lot of randomness in its color pixels, and the background noise has the same kind of randomness. Compression depends on predictable changes in space and time for maximum efficiency.