I'm converting some videos these days, most of them are encoded using wmv2 codec with a resolution of 720p or 1080p and a big (huge) bitrate (between 15 and 30 mbps) if compared to their visual quality. My aim is to keep the same visual quality but reduce their dimensions. I'm using Handbrake to re-encode them in HEVC (h.265).
My questions:
- Is there a (more or less precise) rule of thumb to choose the encoding parameters in order to achieve the smallest file size possible maintaining the same visual quality?
- Is there a (more or less precise) rule of thumb to evaluate whether, once the re-encoding has completed, I have been good at my commitment or there is still the possibility to significantly reduce the dimension keeping the same (or very close) visual quality?
- Is Constant Frame Rate better or worse than Variable Frame Rate in order to achieve my goal?
- Also, see the question at the bottom of the next section.
In your answer please keep in mind that:
I'm not completely new to this argument (although I'm not an expert, otherwise I wouldn't be asking this question...). I know that the technical answer to my question should be: "It depends on your video, if it has a lot of fast moving scenes or not, bla bla bla...". I'm not trying to achieve the best encoding but a reasonable good one at the first try.
Are these suggestions (see the bitrate section) from Google a good rule of thumb or are they youtube-specific?
I know the difference between Constant Rate Factor (average quality) encoding and Constant Bitrate encoding. I always use the former and proceed as follows:
- Encode 30 - 60 seconds (taken more or less from the middle of the video) using CRF (and CFR) with RF 18 (which according to many articles should be 'visually lossless') and compare the quality of the encoded result with the source.
- Then I compute the size of the full video if encoded with the same RF using a simple proportion. If quality and dimension satisfy me I'm done and start the encoding of the entire video. If not (i.e. dimension is still too big) I try to encode the same piece of footage with RF 20, 21 or 23 and choose the one that more satisfy me.
So far I've found that visual quality is quite the same in all cases (RF=18,20,21,23), but dimensions decrease consistently. Which kind of data gets lost during the encode that do not impact on what I see? Is that some kind of data that my screen cannot reproduce (I've got a Late 2011 MacBookPro with 15-inch 1680 x 1050 anti-glare screen) or is something linked to my eye-brain system?
