With a service like that, it helps to have scene detection capabilities as well - so you don't end up having to look at lots of very similar looking thumbnails. So - as opposed to extracting frames at pre-set intervals, extract them when things sufficiently change in the video. That also lets you vary extraction interval between fast action and static talking head-type scenes.
We've done a lot of similar work, will be happy to compare notes.
> We decided not to open source Lumberjack. At the end of the day the whole project is just some business logics around the open source projects I mentioned on this post.
once we fix caching and improve the nginx locations I'll make sure to release under github.com/nytimes but you'll see that there's no much going on over there, just some config files and the lua app. :)
Please don't use the word "retarded" like this. If you suspect you're asking a question with an obvious answer that you've missed, try saying "forgive my ignorance" instead.
There's nothing wrong with asking questions without self-deprecation, but if you're going to self-deprecate, try not to use hurtful language.
Purely curiosity at this point - it was something I wanted for a project a while back that has been retired - does anyone here know of the best way to get a good thumbnail from a video using ffmpeg?
I think back then I was looking to take 5-10 snaps somewhere between the 30-50% mark and then use the largest (I guess in theory more detail - not a blank frame) image
Does anyone have a better way, or even just know how to pull a snapshot at "30%" using ffmpeg? I think at the time I used outside code and ffprobe to figure out where 30% is, there's bound to be a better way than that
If you look at the thumbnails extractor, it calls libav methods that are essentially the core of ffmpeg. One of goal of this service is so journalists and editors can easily play and create thumbnails without command lines and terminals :)
mv4|8 years ago
We've done a lot of similar work, will be happy to compare notes.
https://twitter.com/iTrendTV/status/814543110399229952
contingencies|8 years ago
This should frequently intersect with keyframes for decent/VBR encoded video, if I'm not mistaken.
flavioribeiro|8 years ago
ju-st|8 years ago
NKCSS|8 years ago
:-/
flavioribeiro|8 years ago
kiloreux|8 years ago
FroshKiller|8 years ago
There's nothing wrong with asking questions without self-deprecation, but if you're going to self-deprecate, try not to use hurtful language.
corobo|8 years ago
I think back then I was looking to take 5-10 snaps somewhere between the 30-50% mark and then use the largest (I guess in theory more detail - not a blank frame) image
Does anyone have a better way, or even just know how to pull a snapshot at "30%" using ffmpeg? I think at the time I used outside code and ffprobe to figure out where 30% is, there's bound to be a better way than that
detaro|8 years ago
NicoJuicy|8 years ago
flavioribeiro|8 years ago
nsebban|8 years ago
Over-engineering is sometimes quite fun, though.