On 25/10/13 10:35, Peter B. wrote: > On 10/24/2013 11:52 PM, Ronald S. Bultje wrote: >> Hi, >> >> On Thu, Oct 24, 2013 at 5:29 PM, Peter B. <p...@das-werkstatt.com> wrote: >> >>> On 10/21/2013 12:43 AM, Luca Barbato wrote: >>>> Basically you should make sample files (so 1 frame or 3 frame sample) >>>> that sports specific corner cases or common cases. >>>> >>>> e.g. all the kind of capture you want to make using it. >>> Like, e.g. VHS, DigiBeta, etc? >>> If so, then how many different samples should I provide? >>> The ones I have are usually in yuv422p. >> >> People typically use more standardized existing content, such as: >> http://media.xiph.org/video/derf/ > > Hm... > So, it seems that I didn't yet fully understand the purpose of the > provided samples.
The basic idea is encoding some raw material (like derf collection) using all the features in the encoder so you can test the decoder against the produced samples and make sure it works. > I would also appreciate it, if you could point me somewhere to read up > on what you are actually testing in this case, so I could understand the > choice of samples myself :) E.g. you encode parkjoy using all the ffv1 version .1 features, then you do the same using version .2, then .3 and so on. This way I can make sure the decoder I'm updating is working fine, then I'll do the opposite encoding, making sure the bitstream produced is correct. This kind of samples (small and short) are what we use in FATE to do regression testing. lu _______________________________________________ libav-devel mailing list libav-devel@libav.org https://lists.libav.org/mailman/listinfo/libav-devel
