[FFmpeg-user] Best settings for FFV1

[Etienne Desautels]

Hi,

On 2015-09-17, at 18:45, Dave Rice wrote:
>> On Sep 14, 2015, at 11:56 PM, Etienne Desautels <etienne.desautels at gmail.com> wrote:

>> I will start using FFV1 to archive video for a museum and I will like to know what's the best settings to get smaller files. I will use version 3 with checksums. From the few tests I did I found that those settings were the best for smaller files:
>>   - coder 1
>>   - context 0
>>   - g 1
> 
> I believe using -g 1 would require more size than not. On the other hand this is usually one of the parameters for ffv1 I recommend the most as it increases the resilience of the file which is particularly relevant for archival use.

I had a few exchanges with Peter Bubestinger of the Austrian Mediathek. It's the guy who did the test suite you mention below. Here what he says about GOP:
> GOP (=Group Of Pictures) is a unit very common for lossy encoding.
> If GOP=1, every frame can be decoded standalone, without requiring
> previous or successive frames.
> If GOP>1, frames inside the GOP require the last previous keyframe
> (=I-Frame) for proper decoding of the image.
> 
> There are 2 benefits of GOP=1:
> 1) Video material can be cut at any point. This is useful for
> cropping/editing frame-exact.
> 2) If any errors (bit-/bytewise) happen to the video, the errors do not
> affect more than 1 frame/slice at max.



>> I also tried 2 pass without any success. The final file were exactly the same size as the one created with 1 pass. And the log of the 2 pass was filled only with 0 except for the last number that was something like 843?
> 
> Can you show your process? Have you tried the directions according to the wiki? http://trac.ffmpeg.org/wiki/Encode/FFV1

Here's Peter's answer on the subject:
> Dual-pass mode is currently only working for very short clips.
> It's on my TODO-list to tackle that issue, but it hasn't been solved
> yet, due to lack of time :(

Regards,
Etienne Desautels