Digitizing web videotapes for web distributions.

Easy version

1. Digitize your movie using mediarecorder, following the instructions on the capture tutorial.
   • Split longer videos into into several logical segments of roughly 30 seconds or a minute. (If you just split them on the one minute mark, then they'll break right in the middle of a sentence which is disconcerting. So try to have the segments be aligned with the action.)
   • Use the realtime SGI hardware-dependent compression at this stage. The segment boundaries can be very rough, you'll deal with frame-accurate trimming in the next stage.

2. Import each clip in turn into Premiere, and use its trim capabilities to set exact boundaries. See the Premiere tutorial for pointers. Use QuickTime Composite output, not QuickTime Movie output.

3. Recompress each video segment with device-independent compression using the command-line dmconvert interface. It's much faster to crunch it this way instead of using Premiere.

4. The right place to put your resulting videos is /n/graphics/web/content/docs/videos/. Make yourself a directory to put all the pieces in, and make an index page so that people can download the components separately. It's useful to list both the time and size of each segment. Add pointers to your directory from the main index page if you want the video to be publicly accessible. If your video has an accompanying paper, consider creating crosslinks between the paper page and the video page. Consider also grabbing screenshots of the first frame of each segment as a visual directory, as in this example.

More detail

Recommended first stage digitzing settings

• Size: Use 320x240 (halfsize) frames. Fullsize 640x480 results in files too big to download.
• Compression: use hardware JPEG.

Recommended second stage trimming settings

• Output video: Key difference from main Premiere tutorial directions is that you want to create output that is QuickTime Composite, *not* QuickTime Movie. That means the movie will be created very quickly, with no additional compression. (This is bad when you're making a video to dump to tape, but exactly what you want in this case.)

Recommended third stage compression settings

• Behind the scenes, Premiere actually calling the SGI dmconvert command-line program. You can do it yourself, which I think is considerably faster for major crunching. The command-line syntax has a somewhat high learning curve, but the man page is complete. I (Tamara) spent some time trying to find the right tradeoff between file size and image/sound quality. Here is the command-line that settled on for one of my movies:

  dmconvert -f qt -p audio,rate=44100 -p video,brate=100000,comp=qt_cvid,rate=10,squal=.5,tqual=.5 big.mov svelte.mov
  

  That corresponds to the following settings:
  • File format: QuickTime
  • Audio rate: 44100 HZ
  • Video bitrate: 1000 K/sec
  • Video compression: QuickTime Compact Video == Cinepak
  • Frames per second: 10
  • Spatial Quality: .5
  • Temporal Quality: .5

• I recommend transferring the raw files from moviola to radiance, since dmconvert is all about CPU speed. The transfer time is worth the resulting speedup. Note that Cinepak compression times are still quite long: between 10 and 20 minutes to compress a one-minute segment.

  I have found that the file sizes for movies that are digitized from tape are usually much larger than for movies that are created directly from Premiere. For instance, the H3 video clips of around one minute each range in size from 25MB to 47MB, at quality level .5. In contrast, the direct-from-Premiere Constellation video one-minute clips range from 4.4MB to 11.3MB, at quality level .99 I'm not really clear on why this is so, I don't think it's operator error (although one never knows....).

Additional notes

• The major tradeoff is legibility vs. download time. The biggest win with file size is frame rate. 15 FPS results in files half the size of 30 FPS. I found 10 FPS to an acceptable setting. Temporal quality controls jerkiness, spatial quality controls fuzziness. I found .5 a reasonable setting for both, not much worse than .7. The movie looks pretty scruffy around .2 or .3. To understand the file size tradeoffs, the results from one of my tests was 25MB for spatial and temporal quality of .75 vs. 9MB for quality of .2 spatial and .3 temporal. (Both were 15FPS, and the segment was 47 seconds long.) My final choice was 10FPS, .5 spatial and temporal quality, resulting in a 31MB file.

• You can use dminfo foo.mov to discover the settings used to create most movie files.

• The movie file should have the suffix .mov to be readable on PCs. Don't bother gzipping the movie file, since it can cause problems on PCs: the Cinepak format is highly compressed.

Troubleshooting

• Audio: don't bother tweaking the audio settings, since the amount of space used is negligible compared to the video. (I also could not find a way to tweak the audio bitrate without completely messing up the speed of the audio track.)

Return to Main Video Menu
videolab@graphics.stanford.edu
Last modified: Wed Jun 21 02:14:30 PDT 2000