Personal View site logo
Make sure to join PV on Telegram or Facebook! Perfect to keep up with community on your smartphone.
Please, support PV!
It allows to keep PV going, with more focus towards AI, but keeping be one of the few truly independent places.
WorkHorse
  • 227 Replies sorted by
  • @frullaccia downloading the file...

    the diagonal rain (zebra strips) is a common known issue in mts videos done with hack patches. it is not a problem in the video file, it is the decoder that cannot do a proper decoding. the sotution to this is to transcode the mts file to apple prores using the 5DtoRGB free software, then the problem disappear:

    http://rarevision.com/5dtorgb/

  • @frullaccia I saw the video, great image quality. It is nice to see you got the auto/electronic functions of the lens to work.

    Only thing I would like to improve is the texture quality in shadows areas. It is possible to see that the trees under shadows only have almost one shade of green, some kind of posterization. this is not a datarate problem, increasing datarate or changing the gop tables to balance I/P/B frames size does not solve this. this can only be improved in the scalling tables matrix. the nebula matrix improves this, but nebula frames vary so much along the frames flux so it gets unstable, nebula does not have custom fallback tables. custom fallback tables is which keeps frames constant size along the frames flux to allow stability in high datarates. solution would be to find another scalling matrices which improves this and have good fallback tables to keep frames size constant. I do not know how to create scalling matrices, but probably there is another patch here in PV from which we could extract matrices with better shadows/dark areas textures and good fallback also. the big advantage of flowmotion matrix is that the fallback is perfect, frames are constant size along the fulx so it is pretty stable. to do this flowmotion uses similar matrix in I, P and B tables because fallback is the same for IPB. nebula uses different matrix for IPB so it seems to be impossible to build a custom fallback matrix for nebula which holds the frames size for IPB at same time. the perfect matrices would be similar in IPB tables to make the fallbak holds all frames size constant, but with a better texture in shadows, I think this feature is related to "matrix frequency". improves this would need some help from experts in matrices or to find a patch which already have this to extract.

    http://en.wikipedia.org/wiki/Posterization

    some people think that only 10bit or 12bit codec can handle the great textures in shadows, but the nebula shows that the 8bit codec can handle also. the nebula does not generate more color tones, because this is impossible in the 8bit codec, but it does a more fine dithering and this makes the transition from one shade to another more beautiful than a shade jump. using great matrices, gop3, high datarate and good size balance between IPB in the gop tables can make the 8bit codec to have great shadows textures. the challenge is to find the matrices which have great shadows quality from nebula and the great constant frames size from flowmotion at same time. this would turn GH2 into a low budget Alexa image quality definitely. Maybe this kind of patch already exists in PV but I don't know yet.

    Edit post: I forgot to mention something very special in flowmotion that I perceived. the matrices reduces the cmos pattern noise, and also the chroma noise, and a plus: the flickering in high iso is gone. flowmotion matrices scores 9 out of 10. Sometimes I feel sad to be a pixel peeper to want this better dithering in shadows... but I would like to!!!

  • @apefos At the moment, I doubt there is a patch like that.

    For instance, I noticed a real good chroma quality in the last three @driftwood patches: classix, moon t8, and nebula.

    Speaking about older patches I liked Flowotion and the Cluster X series. Occasionally I also used Canis Majoris Night, Valkyrie 444, and for different purposes Natural Beta 3. Some of these patches is interesting in dark areas. But none is really good as the last three by Driftwood and all your versions. Not to mention about the great improvement with noise (a main problem) and moiré disturbance.

  • Work Horse (7 september version)+ Pixco Speed Booster + Helios 44-2

  • @NikolayM @apefos Right! This patch gives a natural and balanced picture.

  • @NikolayM nice images. it is interesting you ike the versions 07/09, seems that the old versions have something special on them... I will compare with the new versions. @JayUKB liked the 05/09 versions, @frullaccia liked the nebula-workhorse from sometime ago and made me start to think about matrices.

    @frullaccia I am looking into the the patches you said, I downloaded them. I did some research and found that people like the canis for fow light,

    I am trying to understand the hexadecimal numbers in the tables, seems that the frist two digits are for the chroma table and last two digits are for luma table, but can be the opposite, I don't know. from 00 to FF these hexadecimal digitas vary from 0 to 255. it seems they determines how much each pixel will be compressed in the block. there are 48 numbers but I don't know the configuration of the matrix, if it is 3x16, 4x12 or 6x8... (see image) I did some reading in wiki about matrix and I understood the basic principle of DCT, first number is the average value for all, other numbers are how much they vary comparing to the first number.

    I also found this old video with interesting comparisons about textures in dark areas:

    vimeo.com/48488695

    matrix_options_a_b_c.jpg
    433 x 443 - 55K
  • @apefos In fact your older versions (5, 7, 9 september) need a deeper testing, because I like very much the last Workhorse Flowmotion, but I don't know if it is really the best of the bunch.

    Canis Majoris has also an interesting PM version, supposedly made for afternoon light, which seems to be better working than CMajoris Night in HBR mode.

    All these old patches have low chroma fidelity, compared with these two last months series!

  • @frullaccia to my eyes the last uploaded versions here in this topic looks ok. including the new ones in the gspot topic with careful tweaking in gop tables. but people have different preferences.

    to be honest, matrices are a deep sea full of mysteries... so I do not know if I can improve matrices, I am taking it easy, slowly now... there is no other tweaking to do anymore, just matrices coud be changed, so it seems there will be no new versions for a while.

  • @apefos @frullaccia Was busy with video shoot that's why I couldn't test new Patches. Just Tested your new Patch G-Spot on the Highest setting 112mbs / Pal 720p-50fps it's flawless man gives long recording. Max i-frames 276 & P-frames 176 Quality is phenomenal as well. Great work team gonna keep testing it's night time around my way but in the morning I'll test it out outdoors in more extreme situations.

  • @TrackZillas good report. allow me a question: idynamic, image stabilizer and auto focus work in 720p50 in the gspot 112Mbps pal version? the frames size you got was in low light high iso with less detail in image? thanks.

    @frullaccia I am doing some study about matrix, I am starting to understand. but as I am a newbie on this, there is still a long way to go.

  • @apefos i-dynamic off, no image stabilization, no auto focus it was in low light high ISO, and lit room. Will report today with all those turned on.

  • Recently used the Sept 9th Workhorse and The End Patches to make this little video. Thanks to Apefos for his hard work on all of his patches. Vimeo

    Youtube

  • More Sept 9th Workhorse this time in much more low light.

    GH2 Nikon Nikkor 28mm @ f2.8 I had to crank down the shutter a lot and was typically @ 1/40 and sometimes lower so i could milk out any dynamic range I could get. Cinema Profile From memory I think everything is cranked down to -2, except NR, That I set to 0

    In grading the image: There was not a whole lot of ability to adjust the red's and orange color ranges. Part of that is not being able to 100% adjust color temperature when needed, and this was hardly a controlled set.

    However there as a lot of latitude in the highlights.

    The noise was an issue, but I knew that going in considering I was only going to open up to f2.8, would've loved 1.8 or heck .95 but the 28mm is what I had and it went to 2.8...better luck next time. There's a small clip that was done with a cellphone, I'll let you figure that one out.

    Thanks again Apefos, still love this hack.

  • @thpluto how do you compare 'workhorse' vs 'the end nr4' patches?

  • Workhorse sept 9th seems to be more friendly to lumetri color and lighting corrections in post, also the noise patterning for just about everything seems more fine and smoother with the cinema profile at 200 to 1600 iso. The end looks great but I tend to find macroblocking in the falloffs of shadows and as soon as I start pushing colors/LUTs the macroblocks become very prominent. From memory I remember the workhorse noise "dances" more slowly than the End or other hacks for that matter. Noise is fine to me but when it is dancing erratically and has completely different colors than the scene is when it is unusable.

  • Thanks for comparison.

    Did you try "The End NR4 Intra 120" ? It is all intraframe in 24p and 120Mbps datarate and avoids all macroblocking. The other version "The End NR4 Stable" has a stronger noise reduction for more smooth shadows and dark areas, a different approach.

    It is important to use "The End NR4" versions, not just "The End", because "NR4" versions are much more improved.

  • I'll have to try that one again. If I remember correctly the all-I hacks tend to not work in live preview-on camera. For me thats a useful tool to re-check shots to make sure I have them.

  • @apefos can you PM me a link to the patch you are talking about. When I check out The End patch thread there are many iterations, and many updates, and tests. I am not certain which patch is the one you are referencing.

  • the latest versions of NR4 patches are inside the zip file in this post (the zip is 02april2015):

    http://www.personal-view.com/talks/discussion/comment/194204#Comment_194204

  • If you are considering shadows, noise and macroblocking in your tests, there are other patches which can be interesting to try and compare, the "Sharp_05", they are in the zip in the link below, in intra and stable versions:

    http://www.personal-view.com/talks/discussion/comment/205500#Comment_205500

  • I found my old notes, where I tested a ridiculous amount of patches and wrote down the pros and cons. I have tried almost all of your stable patches. I tested sharp for sure but from the gibberish I see here I was seeing more even grain/noise across the image and in the shadows with Workhorse so that threw Sharp out of the running.

    I have in my notes that the best patches I've EVER used have been The End Intra (April), and CanisMajoris (PM).

    If those two had a baby that kept ETC, and Live Preview that would be probably one the best patches ever. I don't have any post production notes for the patches though, so it's hard to say which grade and correct better.

  • @apefos: I decided to test for very 1st time one of the older GH2 patches 'Apocalypse Now' - 6 GOP Nebula '444 Soft'. What I notice is the very big noise at ISO640 or higher.

    So, is it possible to combine the parameters of "Apocalypse Now' - 6 GOP Nebula '444 Soft" and The End Intra "April" (mentioned by @thpluto) in one new patch?

    Also, since you created few "Sharp" patches, would you try your best to make a "Softest" patch?

  • I ended my work with patches.

    You can copy/paste the matrices from one patch to another

  • @apefos: Would you explain in details, please? What exactly to copy/paste exactly? Once done, what do you think the result will be?

  • @apefos: Please, give me explanation or just step-by-step! I also want to try the same procedure with your "apefos_444_02", but really don't know what and how to do. Would you write more info about "apefos_444_02" matrix and its effect on the image? Thanks!