Choose bitrates for multiple encodings

[Lele-brz]

Hi,
I'm doing some test to understand what bitrate to use to perceive a difference between different encodings.
Let's say I want to define 3 categories of bitrate to choose dinamically which one to provide to an internet user depending on his available bandwidth.

I know that this depends on different factors: quality of the source, type of video(cartoon, sport), profile used, resolution...
And after some testing I believe every single video has its own story.

Anyway, I tested with 400, 800, 1500 on a dozen of videos. (bitrate that should allow HTTP progressive download through the internet).
I need to use 2pass to gurantee an Average bitrate, so no CRF.

According to me there's much more improvement between 400 to 800, rather than 800 to 1500.
Does it make sense to you?
How would you choose the different bitrates?

Thanks in advance
Bye

[Lele-brz]

Hi all,
regarding this post, is there some rule of thumb for determing when there's a real visual improvement in increasing the bitrate?
I was considering SSIM but as I read here sometimes it's not considered the most important factor.
I have to say that when watching a 0.94 SSIM and 0.97 you can notice the difference, but on some videos is a huge difference and on some others you can notice just a slightly improvement.

When considering an increase of the bitrate (from 400 to 800) on a large library, what are the parameters to consider to make sure the visual increase in quality is perceived by different people?

Thanks for any tips and idea
Bye

[audyovydeo]

Hi all,
regarding this post, is there some rule of thumb for determing when there's a real visual improvement in increasing the bitrate?


I think there's a million of us banging our heads against this very problem. I haven't found a rule of thumb, I simply postulated that I want my output to have :
SSIM > 0.95
bitrate < 3000kbps

for DV/PAL material. Of course :
1. using crf it means I have to do a couple of test encodes before getting it right.
2. one out of three videos requires 6000kbps to achieve a ssim of 0.94 ....

so : no rule of thumb.

I'd say just target 1500 (I dont know what your resolution is) and save yourself a lot of test time.

cheers
audyovydeo

[Lele-brz]

Hi,
thanks for that.
Yes, I agree that the result depends a lot on the source.
What sometimes I find is that even increasing the bitrate two or three times the result is not that different; especially if the sources are quite grainy (even if they have high bitrate)
At first I thought I was doing something wrong but after some test I really have to say that the quality and kind of source is the most determing factor.

What I'm trying to undestand is if there's some way to automate the decision to provide a higher bitrate encode or not.
Would it make sense to encode at two different bitrates, compare the resulting SSIM and if higher than a threshold consider to have a Higher Bitrate encode?

[audyovydeo]

What I'm trying to undestand is if there's some way to automate the decision to provide a higher bitrate encode or not.
Would it make sense to encode at two different bitrates, compare the resulting SSIM and if higher than a threshold consider to have a Higher Bitrate encode?

We're obviously thinking along the same lines. Last year I was thinking of making a script that encoded, verified ssim and bitrate, and changed settings/reencoded accordingly. I never completed it because :

- I found it impossible to do it in DOS batch (how to get the ssim and bitrate outputed by x264 to a dos env variable ?)
- I found so much variation in source compressibility that my script would have ended up doing many more encodes to achieve my targets, than me doing educated guesses (it must be noted I only use crf, for you it would be easier as you'd feed it a bitrate).

cheers
a/v