Trying some real-world WebM tests

[benwaggoner]

So, there's been a whole lot of talk about WebM's viability a technology and ecosystem, but I've not seen a good head-to-head comparing the latest vpxenc and x264 versions.

I'm starting off with 720p30 "best" perceptual quality, not worring about encoding time. Getting apples-to-apples is tricky, particularly in rate control.

I'm trying to compare a High 4.0 encode to VP8, at 3Mbps with VBR and other constraints appropriate for Level 4.0. Beyond rate control, I want to allow each codec to use its different bitstream and encoder features as best possible.

I want to see what each can do in terms of quality, and what each requires in terms of decoder performance.

For WebM:

Hawaii_720p30_overlay_stereo.avi -o Hawaii_vp8test.webm --best -w 1280 -h 720 --target-bitrate=3000 -p 2 --end-usage=vbr --buf-sz=8333 --maxsection-pct=833 --buf-initial-sz=750 --buf-optimal-sz=8333 --kf-max-dist=240 --auto-alt-ref=1 --lag-in-frames=25 --tune=ssim --threads=4 --fps=30000/1001 --resize-allowed=1 --verbose --psnr

Which the encoder tells me works out to be:

Encoder parameters:
    g_usage                      = 0
    g_threads                    = 4
    g_profile                    = 0
    g_w                          = 1280
    g_h                          = 720
    g_timebase.num               = 1
    g_timebase.den               = 1000
    g_error_resilient            = 0
    g_pass                       = 0
    g_lag_in_frames              = 25
    rc_dropframe_thresh          = 0
    rc_resize_allowed            = 1
    rc_resize_up_thresh          = 60
    rc_resize_down_thresh        = 30
    rc_end_usage                 = 0
    rc_target_bitrate            = 3000
    rc_min_quantizer             = 4
    rc_max_quantizer             = 63
    rc_undershoot_pct            = 100
    rc_overshoot_pct             = 100
    rc_buf_sz                    = 8333
    rc_buf_initial_sz            = 750
    rc_buf_optimal_sz            = 8333
    rc_2pass_vbr_bias_pct        = 50
    rc_2pass_vbr_minsection_pct  = 0
    rc_2pass_vbr_maxsection_pct  = 833
    kf_mode                      = 1
    kf_min_dist                  = 0
    kf_max_dist                  = 240

And for x264:

--level 4 --preset placebo --tune film --pass 3 --bitrate 3000 --keyint 240 --open-gop --ref 9 --slices 1

Thoughts? WebM's documentation is pretty vague; a long VPx tradition continued ...

[smok3]

Html5 video has so many problems, that its just plain boring (that includes slow webm encoding, no full-screen, browsers not being honest about their abilities to play certain format, ect), so what do you plan to achieve?

some older detection test:

(chromium certainly does not play h.264 without additional system decoders.., basically a detection library would have to know every single browser by its name, boring.)

Even if you solve all the mysteries and try to do a webm web video delivery, you will still end-up with some sort of (silly) library (or better: user deciding what would want - switches) that will have to decide what to deliver to each and every browser (flash, html5, quicktime maybe?) - and i'am not even counting different ipods here. So, say the delivery would want to be something else than your usual youtab clone, you are into a huge coding pain (And i'am not saying it is not solvable, i just didn't happen to see any decent solution till now).

[CruNcher]

Didn't Adobe with a big open mouth announced VP8 support for the Flash Player back then

[nm]

Originally Posted by smok3

Html5 video has so many problems, that its just plain boring (that includes slow webm encoding, no full-screen, browsers not being honest about their abilities to play certain format, ect), so what do you plan to achieve?

Ben is doing a pure codec test, ecosystems issues aside, as he said in the very first sentence.

[smok3]

Originally Posted by CruNcher

Didn't Adobe with a big open mouth announced VP8 support for the Flash Player back then

well, that makes even less sense then (if you count out licensing issues), however I'am not claiming anything, just sharing my experience with mentioned format so far, it was a question.