3

I have a thirty seconds 1080p video file that was rendered from Premiere Pro/Media Encoder (target Bitrate 10) and is encoded in H.265. It has 36.6 MB. When i run the command

ffmpeg -i input -c:v libx265 -c:a copy output.mp4

the new file only hast 5.04 MB. mediainfo reveals that the new file has a Bitrate of only 1164 kb/s. Even though one is vastly smaller than the other, both videos look identical (both have a VMAF score above 97 when compared to the uncompressed version).

Does the libx265 smartly choose Bit rate and thus makes the video size that much smaller?

Improve this question
1

2 Answers 2

Reset to default
2

You will have lost data in re-encoding irrespective of perceptual "quality". That is one of the places that the reduction in size has come from. Minor or imperceptible detail will be lost or approximated using algorithms of the compressor.

The major difference though is going to be in the fact that Adobe and ffmpeg have different default quality and bandwidth settings. That "VMAF score" looks similar just tells you that a human probably won't see a difference, not how much data has been lost.

ffmpeg will decode frames and reencode them using whatever its default settings are. Chances are the defaults are aggressive and according to https://trac.ffmpeg.org/wiki/Encode/H.265 it will default to medium quality profile.

The default is medium. The preset determines compression efficiency and therefore affects encoding speed. Valid presets are ultrafast, superfast, veryfast, faster, fast, medium, slow, slower, veryslow, and placebo. Use the slowest preset you have patience for. Ignore placebo as it provides insignificant returns for a significant increase in encoding time.

Slower compression settings will more faithfully preserve the true original quality at the cost of higher file sizes and compression time.

ffmpeg also defaults to a CRF of 28 which will be quite high relative compression

Choose a CRF. CRF affects the quality. The default is 28, and it should visually correspond to libx264 video at CRF 23

It is even possible that ffmpeg has some more advanced compression features enabled by default. Adobe may eschew some encoding approximation features in favour of higher bitrates in order to reduce rendering time, or to preserve quality or other features. It is difficult to know without knowing exactly what settings are in use.

If you want to have a fair comparison between Adobe and ffmpeg outputs then you should render to an uncompressed format, find out exactly what settings Adobe is using and then compress that uncompressed file in both.

Recompressing an already compressed file (decoding and reencoding) will result in generation loss which, while it may not look significant on only one pass, will lose data and quality. After several passes the loss will be significant.

Improve this answer
4
  • Thanks so much for your explanation! Yes I realize re-encoding already compressed video is a bad idea. However I had similar results encoding the uncompressed file using FFmpeg's default settings. Again, thanks a bunch! :D
    – Felix Tröbinger
    Dec 8, 2020 at 10:28
  • @FelixTröbinger one of the problems is that bitrate does not necessarily translate to efficiency or quality. You can easily tune an encoder to "waste" bandwidth by trying too hard to maintain "perfect" quality. Lossy encoding is entirely down to what are considered at time of algorithm design and video encoding to be "acceptable losses". What one encoder may consider to be acceptable may be unacceptable by another and result in bandwidth differences. Significant numbers of differences in choices for default parameters (particularly CRF) results in substantial differences in bandwidth.
    – Mokubai ♦
    Dec 8, 2020 at 10:51
  • For an output meant for human viewing, "true original quality" (whatever that means) is not relevant. That's why services like Netflix rely on perceptual quality metrics to decide bit budget.
    – Gyan
    Dec 8, 2020 at 15:17
  • @Gyan Indeed, the whole point of modern video and audio compression is to discard as much data as possible while being able to either recreate or at least impersonate a faithful representation of the original using whatever means possible. "Quality" is generally in the eye of the beholder and beyond a certain point increasing bitrate will produce nearly no difference in perceived quality. The question though was about why they are different bitrates for a similar perceived quality and the only real answer is because they use different settings to reach their goal.
    – Mokubai ♦
    Dec 8, 2020 at 15:51
1

VMAF is not a good metric at all for very high quality encodes. Even comparing mathematically identical content can sometimes give you vmaf scores as low as around 97.4, while not even vmaf 100 guarantees visual transparency.

Which means VMAF scores tell you almost nothing about the quality of these specific encodes.

At the time, the best visual metric suitable for near transparency video would be a still-image metric like Butteraugli or SSIMULACRA.

Improve this answer

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .