BitJazz
SheerVideo Test Data
Criteria

Let's take a moment to explore why we chose this particular set of test data. After all, it may seem a little odd to use digitized still photographs to test a video codec. But with our perfect-fidelity SheerVideo codec, we're aiming to support the highest-quality video, which is... digitized film. And a digitized film shot is nothing but a series of highly similar digitized still photographs. Now, SheerVideo compresses and decompresses each frame of video as a completely independent keyframe, to ensure that you have instant random access to every frame in the movie, which is essential during production and post-production. Because SheerVideo does intraframe compression only, once you know how well the SheerVideo codec performs for one representative frame of a shot, you know pretty much how well it performs for the whole shot. So giving you precise scores for additional frames in that shot would to tell you a lot less than giving you information about frames from a variety of different shots.

Okay, that explains why we test with digitized still images. But why this particular set of data? The answer is that we're making some pretty extreme claims about SheerVideo's performance here, like encoding 50X faster than the competition (including our own PhotoJazz), yet compressing with 22% more power than PNG. And if we're going to expect you to trust us to publish meaningful and accurate benchmarks about our own products, we feel it's a good idea to make sure that

1. an impartial third party came up with the benchmarks;
2. the benchmarks measure representative real-world data;
3. you or anybody else can easily verify the results.

impartial: As far as impartiality goes, this data set's pedigree is pure. Neither we nor anyone else pursuing image compression had any hand in its production. In fact, it was produced by Kodak for a completely unrelated purpose: commercial stock photography.

real-world: The image compression research community adopted this data set early on because it was one of the first sets of photo-quality images commercially available in digital form, and it includes a wide variety of different subjects, including portrait and landscape, close-ups and vistas, indoor and outdoor, colorful and drab, urban and natural, action and still-life. The one kind of image this set lacks is plain, simple, poster-quality images such as are favored in commercial advertising, business presentations, cartoons, and even certain styles of cinematography. As a result, this test set is skewed toward high detail, and has acquired a reputation of being unusually tough to compress. That's good for us, because most people find that they get better results on their own footage than they would expect from these statistics, and we'd rather have our customers be pleasantly surprised than the opposite.

verifiable: Because these images were published commercially and subsequently standardized as a benchmark, they are widely available. So anyone, including you, can easily verify our results. But what really matters is how well SheerVideo performs on your data, and we encourage you to download SheerVideo and try it out for yourself. Right now. Go ahead.