Personal View site logo
HQ3GOP (3GOP setting optimized for PAL users)
  • Hi guys,

    here's a new 3-GOP setting for PAL users - right now the setting is adjusted for 24H, HBR/25p, FSH/50i and 720p50 (720p50 is 6-GOP).

    edit: V2 is adjusted for all modes.

    It's based on a lot of things I've learnt from @bkmcwd and @driftwood (great guys!!). Actually it's bkmcwd's "Duke Beta 1" with a matrix driftwood once gave to me (back then I called it "high quality"… "HQ"… since it's a sharp matrix). From there on I've fine-tuned the setting…

    This is an "overall" setting producing pretty large I-Frames on highly detailed scenes but at the same time it's also very "sensitive" to motion. I think adjusting the ratio of I- and B-Frames for 1080i-modes (25p/50i/30p/60i) is the most complicated thing to solve with 3GOP settings. In this setting I feel the ratio of I- and B-Frames is just right for HBR/25p and FSH/50i - it shows smooth motion but also a pretty much compression-free image when shooting highly detailed scenes (for instance landscapes with wide depth of field). The setting is also very capable of shooting low key scenes (very good quantization, fine & artifact-free "noise" - attachment #1).

    Regarding codecs I am not at all an "expert" in any way… this setting is purely the result of "detective work". However, the benchmarks and the actual footage look really good.

    With Auto-anything turned off the setting runs stable in 24H, HBR/25p, FSH/50i and 720p50 on SanDisk 95MB/s SDHC 32BG cards (and faster). I've also successfully tested HBR/25p mode on a SanDisk 30MB/s 16GB card (24H requires a faster card). Spanning not yet tested. In camera playback works.

    Check it out!

  • 38 Replies sorted by
  • HQ3GOP versus Valkyrie 444 TZ4.2 in HBR/25p mode...

    First attachment shows a codec torture movie (first frames show static high detail followed by motion). Second attachment shows a low key scene with narrow depth of field (little detail to resolve).

    While both settings produce comparable I-Frame sizes HQ3GOP produces significantly larger B-Frames (under all shooting conditions).

    ValkTZ42_vs_HQ3GOP_chart_motion.jpg
    1494 x 467 - 321K
    ValkTZ42_vs_HQ3GOP_lowkey.jpg
    1494 x 897 - 589K
  • Very interesting. Unfortunately I'm not going to patch my GH2 with HQ3GOP since I have only a couple of "poor" SanDisk 30MB/s cards but, as a user of your Sanity 3.1 HQ settings, I'm glad you're keeping on experimenting. PAL people owe you a lot.

  • @towie Is the used driftwood matrix the one used in moon t7, for softer pictures with over-sharp Panasonic lenses?

  • @rikyxxx

    Thanks!

    Unfortunately I'm not going to patch my GH2 with HQ3GOP since I have only a couple of "poor" SanDisk 30MB/s cards

    As noted above I've successfully tested HQ3GOP in HBR/25 mode on SanDisk 30MB/s 16GB cards on an extremely stressing codec torture movie. I have not yet tested it on SanDisk 30MB/s 8GB cards (since I don't use them anymore).

    @spreeni

    Is the used driftwood matrix the one used in moon t7

    no. It's not present in any of driftwoods settings. He has given it to me 2 years ago or so ...

  • Great work, towi

  • @towi Its a very good setting Thomas, and tantamount to the hard work you have done in research over the years. Excellent stuff.

  • Nick,

    many thanks! In fact I am very pleased with this setting especially in 25p mode. You know very well this setting would never exist without the extremely valuable exchange with you... all the things I've learnt from you (and @bkmcwd). I thank you!

    Now let's bring Spizz T6 to life ...

  • HQ3GOP vs. Duke (Beta 1 !!)

    • First attachment shows a pan over a highly detailed scene (foliage/outdoor daylight) in HBR/25p mode. While HQ3GOP and Duke Beta 1 both produce the same I-Frame sizes HQ3GOP again produces larger B-Frames. JM-SP-Decoder data looks pretty much the same.

    • Second attachment shows a lowkey scene (narrow DOF, little detail to resolve) in HBR/25p mode. Duke Beta 1 shows larger I-Frames, HQ3GOP shows larger B-Frames. JM-SP-Decoder data of HQ3GOP looks a bit more balanced (constant QP on I- and B-Frames).

    • Third attachment shows Duke Beta 1 vs. the soon to be released HQ3GOP.2 in 24H mode (in HQ3GOP.2 24H will be vastly improved ... it's pretty much finished and I think I can post it later today). Same scene as attachment 1 (pan over foliage). Same I-Frame sizes, HQ3GOP again shows larger B-Frames and lower quantization.


    Further plans: in HQ3GOP.3 720p50 will be improved (most likely also 3GOP instead of 6GOP). As soon as 24p and PAL modes are finished I am going to adjust NTSC modes.

    duke_HQ3GOP_HBR.jpg
    1492 x 895 - 574K
    duke_HQ3GOP_HBR_lowkey.jpg
    1494 x 895 - 573K
    duke_HQ3GOP2_24H-preview.jpg
    1491 x 895 - 562K
  • BTW - when I compare HQ3GOP against @bkmcwd's settings it's absolutely not my intention to devalue bkmcwd's settings in any way. Quite the opposite we have been working closely together over months to improve and stabilze the settings and bkmcwd's settings are by far the best for HBR-mode in 3GOP (IMHO). Also HQ3GOP is based on bkmcwd's brilliant work (actually it's a modified bkmcwd-setting, just with a different matrix). I just wanted to make this clear in order to avoid ambiguities!

  • @towi

    As noted above I've successfully tested HQ3GOP in HBR/25 mode on SanDisk 30MB/s 16GB cards on an extremely stressing codec torture movie.

    I got it, but I need both 24p and 25p to be reliable on my 30mbs cards. I'll check 24L (maybe HQ3GOP.2 one) when I have time to. Thanks for this new patch.

  • Here is HQ3GOP V2

    The setting is now adjusted for PAL and NTSC modes: 24H, HBR/25p, FSH/50i, HBR/30p, FSH/60i, SH 720p50, SH 720p60.

    25p/50i is the same as in V1 ... all other modes have been improved.

    I have not yet tested the lower modes (24L, FH, H).

    All 1080-modes (24H, HBR/25p and HBR/30p) and 720p50 survived my codec torture movie even on SanDisk SDHC 30MB/s 16GB cards. 720p60 requires SanDisk SDHC 95MB/s 32GB cards or faster.

    In camera playback works. Spanning not yet tested.


    24H_codectorture.jpg
    1162 x 465 - 292K
    24H_foliage.jpg
    1182 x 473 - 297K
    HBR_25p_codectorture.jpg
    1259 x 469 - 309K
    HBR_30p_codectorture.jpg
    1193 x 468 - 302K
    720p50_codectorture.jpg
    1182 x 471 - 292K
    720p60_codectorture.jpg
    1174 x 471 - 297K
    30p_lowkey.jpg
    1207 x 468 - 292K
    HQ3GOP_V2.zip
    1K
  • @bkmcwd just posted my codec torture movie (thanks, mate!): http://personal-view.com/talks/discussion/comment/137209#Comment_137209

    that is the movie the respective "codec torture" screenshots from above refer to.

  • I like that movie, movie!

  • @Fool4UAnyway

    I like that movie

    I hate it ... it's absolutely merciless and unbribable :-)

  • For your reference here is HQ3GOP compared to several 3GOP settings in HBR/25p and 24H.

    The settings are Duke Beta 4, FlowMotion 2.02, Spizz Trial 4 (modified by me for better 25p), Valkyrie 444 TZ2, Valkyrie 444 TZ4.2.

    The scene is a static texture with narrow DOF (still containing quite some fine details to resolve... see first attachment).

    note: FlowMotion 2.02 actually runs out of competition in HBR/25p mode since it is not really stable in this mode (card speed error even on SanDisk SDXC 95MB/s 64GB and/or it shows mosquito noise on very demanding scenes).

    also note on the HQ3GOP HBR/25p chart I was walking around ... this is why the stream shows some waving from frame nr. 200 and on...

    tex_scene.jpg
    800 x 450 - 161K
    HQ3GOP_Duke_HBR25p.jpg
    1233 x 934 - 615K
    HQ3GOP_FloMo2_HBR25p.jpg
    1262 x 937 - 603K
    HQ3GOP_SpizzT42_HBR25p.jpg
    1241 x 938 - 620K
    HQ3GOP_ValkTZ2_HBR25p.jpg
    1316 x 937 - 609K
    HQ3GOP_ValkTZ4_HBR25p.jpg
    1282 x 937 - 606K
    HQ3GOP_Duke_24H.jpg
    1277 x 931 - 575K
    HQ3GOP_FloMo2_24H.jpg
    1232 x 936 - 572K
    HQ3GOP_SpizzT42_24H.jpg
    1321 x 936 - 580K
    HQ3GOP_ValkTZ2_24H.jpg
    1237 x 940 - 586K
    HQ3GOP_ValkTZ4_24H.jpg
    1239 x 936 - 583K
  • HQ3GOP V3

    • 720p50 improved

    • 720p60 improved

    HQ3GOP_v3.zip
    1K
  • @Towi, I'm glad to see that you you're developing these, your own, settings. I only have a GF2, for now, but will give your latest ver.3 a workout in 720p_50 mode.

  • @trollwi

    FlowMotion 2.02 actually runs out of competition in HBR/25p mode since it is not really stable in this mode

    Same old bullshit, eh?

    /thread

  • The last time I've mentioned that FloMo2.02 is not stable in HBR/25p mode you've attacked me and you were asking for a proof.

    This time I've prepared one. Here is FloMo in Action on SanDisk SDXC 95MB/s 64GB shooting my Codec Torture Movie: https://dl.dropboxusercontent.com/u/99380599/GH2_FloMo_SDXC64GB.mov And this isn't even overexposed (overexposed torture chart is more prone to crash, of course).

    I've just made another one - especially for you. FloMo mosquito noise, original MTS file (100% reproducible all the time): https://dl.dropboxusercontent.com/u/99380599/mosquito_noise.zip

    Back then FlowMotion was the reason I've started to create my own codec torture movie. Fast pans suddenly stopping on a highly detailed scene or someone passing in front of the lens while shooting a highly detailed scene caused either mosquito noise or card speed error.

    HQ3GOP and also all of @bkmcwd's 3GOP settings do not show mosquito noise in HBR/25p mode. The way I shoot (all M… no auto-anything) they also don't produce card speed errors… not even on my codec torture movie.

  • Hi Towi, I tried the beta2 a few days ago and wanted to shoot some test recordings in FSH (PAL). But unfortunately I had a freeze immediately after start record (after 1 or two seconds). This happened after 10 recordings which were o.k. - so I thought this has to do with the early beta state. I will try your settings again later - it's good to know, that in this setting the PAL side is in focus of development.

  • @spreeni

    at the moment I don't have the time to adjust HQ3GOP further.

    In the meantime you may try these modifications:

    • Video Bitrate FSH/SH=68000000

    • 1080i Top Setting=67741

    • 1080i Bottom Setting=60967

    • 720p Top Setting=67737

    • 720p Bottom Setting=60963

    Please also try Spizz T6 in 25p/50i mode and let us know your findings -thanks!

    Spizz T6 Info: http://www.personal-view.com/talks/discussion/comment/138134#Comment_138134

    setting: http://www.personal-view.com/talks/discussion/comment/134530#Comment_134530

  • @spreeni

    could you please try this preliminary version of Spizz T6?

    Many thanks!

    EDIT: please use the third attachment

    Cluster-X Spizz T6 towi retuned V2.zip
    2K
    Cluster-X Spizz T6 towi retuned v2.zip
    2K
    Cluster-X Spizz T6 towi retuned final.zip
    3K
  • Update: HQ3GOP_V6 (V4 and V5 not posted... so V6 is the successor of V3)

    Completely revised version. There are 2 settings: one for PAL modes, one for NTSC modes.

    All modes have been tested and survived my codec torture movie (this one: http://personal-view.com/talks/discussion/comment/137209#Comment_137209 ... but uncompressed, of course) for several minutes. However, since I myself shoot 24p and PAL modes I've done more tests in 25p/50i than in 30p/60i. In HBR/25p mode the setting spanned on my codec torture movie on SanDisk 95MB/s SDHC 32GB cards. 30p/60i also runs stable but reliable spanning needs further tweaking (however, I think it spans on less demanding scenes).

    Please check it out and let me know your findings.

    HQ3GOP_V6.zip
    2K
  • While HQ3GOP V6 in 25p mode does not span more than 2 files on my codec torture movie it does span on demanding real world scenes. Today I've successfully recorded 45 minutes (6 files spanned) of a highly detailed "landscape" scene. Average bitrate utilzed around 64M.

  • Update: HQ3GOP V7

    Improvements on all modes (modified Scaling Tables on all modes).

    In HBR/25p, on highly detailed scenes, this setting shows the largest I-Frames of any (stable) setting available (above 800K). At the same time it's very sensitive to motion... so B-Frames go pretty large when required. Also Streamparser's JM-SP decoder values look very good.

    HBR/25p survived my Codec Torture Movie for around 15 minutes even on cl.10 SanDisk 30MB/s SDHC 16 GB cards (2 files spanned). This is why I fully expect reliable spanning even on demanding real world scenes. All other modes also survived my Codec Torture Movie for several minutes on SanDisk 30MB/s SDHC 16 GB cards (only 720p60 not yet tested on this card... but successfully tested on SanDisk 95MB/s SDHC 32GB cards).

    24H_codectorture_Frames5.PNG
    746 x 412 - 51K
    HQ3GOP7_ctm_loopV2_Frames2.PNG
    746 x 412 - 47K
    HQ3GOP_V7.zip
    1K