GSpot FlowMotion - versions 17 september 2014 - patches for GH2 hack

separate versions for NTCS and PAL

Best balance between I, P and B frames size to get both noise and details quality in the same patch for 720p. Also tweaked for HBR. 24p already ok from previous tests.

for NTSC:

versions A to F - for lower speed cards, with gspot configuration, but the lower datarate can hurt quality

version G - the gspot, works with the Sandisk 45Mbps card, the minimum datarate to get great noise and details in same patch for 720p60.

versions H to J - for higher speed cards, with gspot configuration, the higher datarate can increase quality even more, but there are chances for stop recording.

for PAL:

versions A to D - for lower speed cards, with gspot configuration, but the lower datarate can hurt quality

version E - the gspot, works with the Sandisk 45Mbps card, the minimum datarate to get great noise and details in same patch for 720p50

versions F to J - for higher speed cards, with gspot configuration, the higher datarate can increase quality even more, but there are chances for stop recording. Give a try on setG PAL, similar datarate of NTSC G version, should work ok.

Matrix:

FlowMotion Scalling tables and Deblocking tables deviler a constant frames size along the frame flux. This allows to increase the datarate without the risk of stop recording or camera freeze. The fallback tables are very well calibrated with the I, P and B tables to keep the frames size constant. This feature in Flow Motion makes the patch to work ok with the frame limit and frame buffer. And this constant frames size along the flux is what made possible to find the perfect balance between the I frames size and the P / B frames size to get both noise/texture and details quality in the same patch, the gspot.

GSpot Nebula - 23 september 2014

different versions for NTSC and PAL

tweaked matrices, custom fallback tables

just one version setg 96Mbps for 45Mbps cards and above

after feedbacks from users and testers I can try to make it work for higher datarates, (do not try to raise datarate in these versions, it will get unstable)

Enjoy!

Ok. FSH and H pal modes are working very good.

@frullaccia thanks for feedback.

It's somethink like in the Star Wars IV when Luke SkyWalker got his hands in the lightsaber Obi Wan Kenobi gave to him. I am just a humble padawan learning slowly from a Jedi Master...

Matrices are where the image texture magic happens... and controls the homogeneous frames flux...

@zsound thanks for feedback. appreciate you follow the work. I will take a look in the smooth cinema.

after the hard work to learn how things works in PTool and in the ini file (it was a deep dive into it), now I am doing things in paralel with my other tasks. I will experiment the other matrices. just a little bit slower work to not prejudice my other tasks. I think this is good also for people to experiment one new matrix version at a time.

I also need to perceive if there is the need of changing quantizer values when changing the matrix, things worked ok on nebula version with same quantizer values from flowmotion versions, but not sure yet if I can keep same quantizer values for all types of matrix.

I will upload versions for lower speed cards as I did before, but I think the way to get the best benefits from the beautiful image texture it is good idea to use a 95Mbps card. Or at least the 45Mbps card which I own and works ok with the setG.ini versions. The Sandisk 32GB SDHC 95Mbps is not so much expensive these days, I am considering to get one to test the setH.ini, I and J versions with the higher datarates.

@apefos
As a user of Flowmotion 2.02 I shoot documentaries (mostly 720p50) with long takes, so spanning is important for me. Is it correct that your patch is a further development of LPowell's patch?

1. I'm not sure which is the newest development of your patch, is it GSpot FM or GSpot Nebula, or are they completely different?

2. What are the benefits of GSpot FM compared to FM 2.02? Is there, for example, less aliasing in 720p50 or less noise?

3. The good thing in FM 2.02 is, that all other modes work stable (also with spanning) when using 95MB/s Sandisk card, is GSpot FM also that stable in all modes or is your focus of development more to 720p50(60)?

@spreeni

best thing to do is to try it and compare, seeing the same image with different patches will bring you the answers you need.

1-at this moment, matrices are from FlowMotion or Nebula, so noise and texture are different. GSpot means that the frames size are balanced to get good noise, good textures and good details in low iso at same time. this is difficult to get in 720p and it is the main achievement of gspot. the other modes also works this way. datarate, gop size, gop tables, frame limit and frame buffers are tweaked in gspot to get this, and then I change the matrix, The flowmotion matrix are same as original, the nebula matrix have tweaking in gspot version, sometimes there is the need of adjusting the original matrix to make it work good in gspot.

2-the goal in GSpot is to show the texture in all areas of the image without texture mud. to show the high iso noise, not hide it. and good low iso details at same time. It just try to show what the sensor outputs and what the matrix does. To do this it increases the datarate, use low gop, and balance I-P/B frames size, to extract the maximum from the camera in the encoded video.

I like to use NR-2 in camera, but you can try increasing noise reduction in camera to perceive the best results for your purposes, experiment the different noise reduction amounts in the GH2.

3-I do not have the answer for this question. My card is the 45Mbps so I cannot answer. The goal here is the image quality, not spanning. some people reports that got spanning using my settings, but my main goal is image quality for short shoots 5 - 8 minutes. mainly for cinema production. The GSpot Flowmotion versions have datarate options from 48 to 112Mbps so probably one of them will be the best for you to get quality and stability. The Nebula versions just have one datarate option at this moment, when I have time I will do it in other datarates.

@zsound @frullaccia

I am advancing other tasks which was delayed due to efforts in the patches... So It will take a while to try other matrices and upload new versions.

@zsound Thanks for pointing to the 'Space' patch. I saw this one before but I'm mainly focused on shorter GOP patches.

@apefos Thanks for the detailed explanations. Sounds very promising, so I will do some tests.

@spreeni thanks for the good reports. interesting comparison.

I will upload new versions as soon as I have time.

@spreeni If you got 3 or 5 4GB files in 720p you can consider it pretty good, much more than expected. It is very difficult to get spanning working in 720p at high datarates in gop3, so your results surprises me. I don't think that changing settings in the ini file will do it better, it is already in the limits of stability. One way I think would deliver better spanning would be using a lower datarate version, try the set E version for PAL, it has the minimum datarate for gspot good noise/details and the lower datarate maybe can help spanning to work better. But as I said 3 or 5 4GB files is pretty good, I was expecting only one 4GB file.

@frullaccia next post I will upload the Nebula versions in all datarates.

@apefos Thank you for the explanations. I just tried the 72 Mbit settings and spanning in 720p50 seems to work as stable as in original Flowmotion. Now I will check quality compared to the higher bitrates. A stable 3-GOP value in 720p50 mode seems to be very attractive :-)

Just as an idea: How about using high bitrates in FSH/SH and lower bitrates (for longer recordings) in SH/H. So I could easily switch between safer settings for filming concerts or interviews and high bitrates for the rest. I think LPowell did it that way, would it be possible for GSpot, too?

FM: 720p50 always passed the death charts. 720p60 didn't.

That's right, but FM uses 6-GOP for 720p50, GSpot works with 3-GOP in that mode. As far as I know all other 3-GOP-patches use 6-GOPS in 720p, too. I'm still testing if this makes a difference in quality.

@frullaccia good news, curious to know about your feedbacks.

@spreeni in my tests with the GH2 I found that keep the same datarate in the SH and in the H slots helps to improve stability. I did tests with different datarates, higher in SH and lower in H, but then recording in H freezes the camera, when both have the same datarate the camera does not freeze anymore. I do not know the reasons for this, it just happens... so I perceived that keep the same datarate in both slots was better for stability, this is the reason I did 10 versions for PAL and 10 versions for NTSC, starting at 48Mbps and ending at 112Mbps from setA to setJ. If you want to try, change the datarate in the H slot to a lower value in the ini file and also the H top limits near the end of the ini file and do some tests.

@driftwood thanks for the feedback, appreciate it. please read the elaboration below:

In my tests I perceived that the number of fps is something that makes the stability difficult. In a first moment I thought that if the datarate was the same, so 1080p24 and 720p60 would work ok, but it is not like that. when increasing the fps to 50 and 60, even with a lower datarate and even 720p being smaller frames, the camera becomes more prone to stop recording and/or freeze. it seems that the calculations to pack more frames stress the camera, so 720p60 is more difficult than the 720p50.

there are two kinds of fails: 1-when the camera just stop recording, save the file and go back to stand by mode, so you can use the saved file and can start recording again: if this happens I believe it is a problem of low speed card or datarate is high to the camera to handle, so maybe a faster card will help, or load a lower datarate version will help. 2-when the camera freezes and needs a battery pull: if this happens I believe there are bad settings in the ini file. In all my tests with the GSpot I got no freeze, I tested the 112Mbps version in the 45Mbps card and it recorded some seconds without camera freeze, so I am wondering that the settings are ok. the frame limit and frame buffer are already very high which I perceived to work better than lower values and increases proportionally to the datarate following measurements in streamparser, but maybe there is something that can be done... suggestions?

so, in 720p60, which is the most difficult mode for stability due to the number of fps, if the camera does not freeze in gop3 high datarate it is something already good. I got stop recording in 720p60 in the death charts also after some minutes, but I did not get camera freeze. depending on the datarate version I load to camera I can record more time or less time and I perceived that the 96Mbps version setG for NTSC 720p60 works ok for recording some minutes in real world situations. so in the real world shoots it is working fine for cinema production with short takes, because real world shoots are not so heavy in terms of details as the death chart. Maybe the lower datarate versions can record more time or a faster card could be a good idea. (my tests was with the sandisk 45Mbps)

I agree that it would be much better to record a full 4GB file in the death charts in 720p60, I did not test the lower datarate versions in long time recording, I just tested the setG and it gives me some minutes without freeze. But there is something we need to consider: the GSpot try to reach the limits of the camera, the highest possible quality in 720p50 and 720p60, so maybe the setG versions are already touching the limits of the camera hardware and there is nothing that can be done in the ini file anymore... maybe... So using a lower datarate version can be useful for long time recording and a high datarate version useful for maximum quality in short shoots. but maybe there is something which can be done... maybe...

I would like to ask: which card did you use? was it the 95Mbps sandisk? which ini file version you tested? was it the setG NTSC or other? did you test in the still chart or in the movie chart? which still chart you used, was it the SDC4? did you get camera freeze or just stop recording? how many time you can record before stop? did you compare the FM versions with the Nebula versions? these answers are important for me to perceive what happens.

suggestions to improve are welcome!

thanks.