Git G3 Duo Firmware Mod

user777

Member
Joined
Jan 29, 2019
Messages
44
Reaction score
6
Location
Finland
Country
Finland
Dash Cam
GitUp Git2P
@hc_1982 first of all, I don't know what your normal job is but let me say that if it doesn't have to do with imaging, videography and software you are wasting your talents! I must admit that delving deeply into the science of curves took some effort as I am nowhere near an expert as you are, but I do agree with your suggestions.

Today I took some more test videos with the test FW while hiking but this time I had all settings set to normal. Normal colours, mid sharpness, mid contrast, EV 0 BUT highest possible resolution. The reason why I did this was because I watched the previous 1080p test videos on my 4K TV and they didn't look so great.
Yes, they looked much better than the stock videos but they were oversharpened, a bit too saturated etc. I think that my 1080p Mac monitor shows the videos a bit more neutral, so the colorful mode is ok. But on a big 4K TV things are different.

In any case I was pleasantly surprised with the results. Very nice sharpness, balanced colours (if only a tad to the neutral side) and slightly lowish contrast.
What I am trying to say is that my latest tests show you are correct in your curves assessment. I was basing my conclusions on the Mac monitor, but maybe I should also review them on the big screen too.
Right now I think that the mid contrast of the test FW together with Normal colour mode tends a bit to the low side, so we can increase it slightly. Due to work I am not able to do more frequent tests, so I suggest you take also a few sample videos with the various contrast settings and lets see how we can tweak the curves more.
 
OP
H

hc_1982

Member
Joined
Aug 19, 2018
Messages
95
Reaction score
63
Country
Ukraine
Hello @user777 , actually, the only thing I knew about photo/video before I bought git2p, was how to turn camera on/off, point and shoot. But I have been disappointed right after start using it, especially when used indoors.

1080p footage looks fine on low-res displays (1360x768 and lower), and not so good even on 1080p display, so I guess, it should be even worse on 4K TV. Also, it seems that each display shows video a bit differently, as different gamma/contrast/saturation settings are used, and according to my experience, it may look perfect on one display and absolutely horrible on another one, so I'm trying to test video on every device I can reach after major changes. Also, lot of video players are not able to handle video format properly used by these cameras and may clip highlights and shadows, which results in excessive saturation and contrast of displayed video (B/W effect is for such cases).

For higher resolution video on git3, I guess, optimal one would be 1440p30, as high bitrate is possible here and less rolling shutter due to higher readout rates compared to 1620p/2160p.

I have prepared new curves in order to handle overexposure better and improve contrast, and updated color correction settings for reducing yellowish tint of video. Also, some filtering settings have been updated, so 1080p50/60 looks a bit better now as for me. Please check it out, when you have chance:

 

user777

Member
Joined
Jan 29, 2019
Messages
44
Reaction score
6
Location
Finland
Country
Finland
Dash Cam
GitUp Git2P
Hi @hc_1982! Thanks again for all your efforts. I will test the new FW and report back as soon as I can.
In the meantime I have a little something for you that you may like.

Recently I put the latest modified firmwares on all my cameras (Git2p, G3, F1) and I shot a small test video with all of them to compare the colours, saturation, curves etc.


The Git2p seems to be a bit cold while the F1 a bit too vivid for my taste. But IMHO the G3 has the most pleasing and balanced output. Punchy colours without becoming too much.

Note that I was using your last test FW with all settings stock. So you should be proud of the work you are doing; I think it is really paying off.
 
OP
H

hc_1982

Member
Joined
Aug 19, 2018
Messages
95
Reaction score
63
Country
Ukraine
Hello @user777, thank you for your testings and review! Please note, that new test mod (test5) has different curves and color correction settings, so image could be quite different with it. I hope, that it would be possible to make colors better, as several improvements have been done to script which calculates LUT for 3D color correction, but it still requires some tuning, and probably colors could be oversaturated now.

I guess, F1 hardware is more powerful, compared to Git2P and Git3, as it has a12s75 Ambarella chipset, IMX317 sensor and more RAM. It can handle 4K30 video, and has more sophisticated software, so, technically it should be capable of producing much better video output, while Novatek chipsets used in Git2P/Git3, is quite outdated and software used in it is quite simplistic, although this minimalism makes modding of firmware much easier too.
 

user777

Member
Joined
Jan 29, 2019
Messages
44
Reaction score
6
Location
Finland
Country
Finland
Dash Cam
GitUp Git2P
Good morning @hc_1982

I just wanted to make a quick note, as I haven't had the change to review your test5 FW, but I think we need to step back for a minute and address another issue: rolling shutter, or jello effect.

Last summer I mounted my Git2p on my motorcycle and traveled for about 5000 km across Europe. I took many videos on stock FW @1080p30 and I never noticed any serious jello effect in them. However, just some time ago I used the F1 and the G3 on short rides around with modded FWs and noticed immediatelly pronounced and very ugly jello effects.

Initially I suspected the mount. Obviously it plays a role. If there are no vibrations (like in a car) then there is no/very little jello effect.
However, the mount on the bike is exactly the same as it was last summer. That should be no reason that the Git2p has no jello on this bracket while the F1 and G3 do.
So I turned my attention to the settings/ firmware.

Yesterday I managed to take a moment and go out with the G3 filming 1080p60 and then 1440p30 and the results were very surprising.
At 1080p60 fps the jello effect was very limited; still there but not too bad.
However, when I switched to 1440p30 all hell broke loose! The jello effect was so bad that it made the video basically unwatcheable.
Nothing else was touched other than the settings. Positioon the same, bracket the same, road the same.

So I was wondering if the modded FW with its higher bitrates pushes the processor or the imaging chip a bit too much, therfore struggling to keep up with fast moving objects? I read about the rolling effect and it is due to the way all modern sensors read/write information from top to bottom in comparison to older processors which read/wrote all the info at once.

I put back the stock FW for now and I will try to get a few more tests done, but I was wondering about the possible connection to the much better IQ of the modded firmware and the additional strain it might be placing on the processor of the camera.
 

Tractor

Well-Known Member
Joined
Aug 24, 2018
Messages
442
Reaction score
415
Location
Dover, Kent
Country
United Kingdom
Dash Cam
DCTeam A119 V3 - A119 V2 X 2 - GitUp G3 Duo - Yi DashCam
Quite normal jittery video when filmed from an unstable vibrating mount. . But without a stable not vibrating mount you will still suffer Jello.
 
Last edited:
OP
H

hc_1982

Member
Joined
Aug 19, 2018
Messages
95
Reaction score
63
Country
Ukraine
Hi @user777,

I believe, jello-effect is produced by rolling shutter, and depends on sensor readout timings. The faster frame is obtained from sensor, the less jello will be noticeable. One sensor mode is used in git2p original firmware for both 1080p30 and 1080p60, while different sensor mode is used for 1080p30 in modded one, the same for 1440p, which produces only 30fps. So, difference for 1080p30 between mod and stock firmware would be obvious there in terms of jello effect. But things should be different for mod for Git3, as original mode used for 1080p30/1440p are able to produce ~40fps, while different sensor mode is used in fw mod for these resolutions, which is capable of 48fps, so I guess, it should produce less jello compared to stock firmware. Also, technically it should be possible to achieve the same level of jello for 1080p30/1440p on git3, as observed with 1080p60 by switching sensor mode, as only difference between them for mod right now is color depth (12 vs 10 bits, and I cannot see much difference in video), and read-out times. Also, I believe, enabling EIS should reduce jello effect, as it is supposed to do some rolling shutter correction among other things.

Current modded firmware for git3 does not push camera processor harder compared to the stock one, and mostly difference is workload for microSD controller and card itself, which have to write 2 times more now, although higher load will be for 1080p60 in next version of mod, as I have changed some settings for it recently in order to improve resolution. So, if you have not tested "test5" mod, please wait for "test6" which will be ready for tests soon.

Also, please let me know, about your findings regarding jello effect with stock fw.
 

user777

Member
Joined
Jan 29, 2019
Messages
44
Reaction score
6
Location
Finland
Country
Finland
Dash Cam
GitUp Git2P
Quite normal jittery video when filmed from an unstable vibrating mount. . But without a stable not vibrating mount you will still suffer Jello.
I think you missed my point completely. With the same mount, on the same motorcycle, the Git2p produced very good jello-free video over thousands of kilometers, whereas the F1 and G3 are suffering by comparison. How do you explain this?

The ONLY difference was (apart from the cameras) the firmware. The Git2p was running the stock FW, whereas the F1 and the G3 are running the modded FWs. Mount, position and everything else was identical.
How can you simply shrug it off as a problem of the mount?
 

Tractor

Well-Known Member
Joined
Aug 24, 2018
Messages
442
Reaction score
415
Location
Dover, Kent
Country
United Kingdom
Dash Cam
DCTeam A119 V3 - A119 V2 X 2 - GitUp G3 Duo - Yi DashCam
Try Turning off EIS as it maybe over compensating.
 

user777

Member
Joined
Jan 29, 2019
Messages
44
Reaction score
6
Location
Finland
Country
Finland
Dash Cam
GitUp Git2P
Hi @user777,

I believe, jello-effect is produced by rolling shutter, and depends on sensor readout timings. The faster frame is obtained from sensor, the less jello will be noticeable. One sensor mode is used in git2p original firmware for both 1080p30 and 1080p60, while different sensor mode is used for 1080p30 in modded one, the same for 1440p, which produces only 30fps. So, difference for 1080p30 between mod and stock firmware would be obvious there in terms of jello effect. But things should be different for mod for Git3, as original mode used for 1080p30/1440p are able to produce ~40fps, while different sensor mode is used in fw mod for these resolutions, which is capable of 48fps, so I guess, it should produce less jello compared to stock firmware. Also, technically it should be possible to achieve the same level of jello for 1080p30/1440p on git3, as observed with 1080p60 by switching sensor mode, as only difference between them for mod right now is color depth (12 vs 10 bits, and I cannot see much difference in video), and read-out times. Also, I believe, enabling EIS should reduce jello effect, as it is supposed to do some rolling shutter correction among other things.

Current modded firmware for git3 does not push camera processor harder compared to the stock one, and mostly difference is workload for microSD controller and card itself, which have to write 2 times more now, although higher load will be for 1080p60 in next version of mod, as I have changed some settings for it recently in order to improve resolution. So, if you have not tested "test5" mod, please wait for "test6" which will be ready for tests soon.

Also, please let me know, about your findings regarding jello effect with stock fw.
@hc_1982, you just confirmed my suspicion and thanks again for your detailed technical insight! I don't mean this as flattery, but to have a person of your technical knowledge on this team really makes an incredible difference.After reading an excellent article on the rolling shutter problem and how modern sensors do their readouts line by line, I started suspecting the speed of the sensor's readouts as a possible culprit.

Just yesterday I was seeing some short clips by a biker friend who has a modern GoPro fixed on his bike with the stock GoPro mount. The footage was rock solid! Absolutely no rolling shutter and clear footage. Since we know that the stock GoPro has a high sensor readout/bitrate capability that could go some way towards explaining the lack of jello.

In 10 days I will start on another cross-Europe journey on my bike and I need the camera to be working ok by then. For now I will give the standard FW a go on 1080p30, p60 and then on 1440p30 and see what we get.

By the way I was thinking if it would be possible to combine the greatly enhanced photo abilities from your FW with the stock video section of the OEM FW. Can this be done even by me?
 

Tractor

Well-Known Member
Joined
Aug 24, 2018
Messages
442
Reaction score
415
Location
Dover, Kent
Country
United Kingdom
Dash Cam
DCTeam A119 V3 - A119 V2 X 2 - GitUp G3 Duo - Yi DashCam
Rolling shutter is an interlace and not a progressive problem.
 
OP
H

hc_1982

Member
Joined
Aug 19, 2018
Messages
95
Reaction score
63
Country
Ukraine
@user777, current mod does not have much changes specific for photo, as far as I remember. The only changes have been done for it, are related to intermediate resolution of image, which was reduced from 4000x3000 to 2880x2160 originally (shared with 2160p24 video), and I have changed it to full resolution without resize, but not sure, if it actually had any noticeable effect. Also, some compression settings have been changed to produce .jpeg faster and with bigger file size. Also, I have reduced noise reduction settings for photos in latest test mod which will be published soon, so image seems to be less blurry/smoothed now.

For git3 we can achieve at least the same level of jello effect stock fw has, so, I believe, no much need for it. But I think, ntk firmware modding could be done by anyone willing to fiddle with it and has some mips asm and C basic knowledge and some reverse engineering practice. So, it depends on your on programming/disassembling skills or something like this. In case you have no such previous experience, it would be quite hard to start with modding in 10 days.

As I heard, latest Hero 6/7 Black are using some specific chip for image processing developed by Socionext, which is very good in electronic image stabilization and rolling shutter correction, and better sensor, as I can judge by public specs for IMX277, it should be able to perform readout up to 2 times faster for 10-bit video modes (~90fps for fullres, and ~120fps for 4K2K).

Ok, let me know about results of your testing with stock fw, and I will check what can be improved for git3 mod 1440p/1080p30 to reduce jello.
 

user777

Member
Joined
Jan 29, 2019
Messages
44
Reaction score
6
Location
Finland
Country
Finland
Dash Cam
GitUp Git2P
@hc_1982, I don't have any experience in C programming so no point in trying anything at the moment.
Question: will you reduce the video mode to 10-bit in your coming test6 FW to help with the rolling shutter?
Also is the stock FW for 1080p using 10 or 12 bit mode?
 
OP
H

hc_1982

Member
Joined
Aug 19, 2018
Messages
95
Reaction score
63
Country
Ukraine
@user777, stock firmware is using 10-bit sensor modes for each video resolution, as far as I can tell at the moment. I have added another sensor mode (4096x2160, 48fps, 12-bit) which was not originally supported by firmware and switched 1080p30, 1080p48, 1080p25, 1440p30 and 1440p25 to this mode (25 and 48 fps video resolutions are not available in stock firmware too). So, first of all I will check timings for this new sensor mode and what timings are possible for these framerates with this mode and the one used for 1080p60 (4096x2160, 60fps, 10-bit).
Stock FS for 1080p30 is using sensor mode with dimensions 4000x2250, 40fps, 10-bit, so I guess, it should be possible to achieve up to 20% faster readout speed even with current sensor mode, but jello effect strength probably depends on actual timings used for single horizontal line readout, which is a bit less for 10-bit modes, and minimal value depends on sensor mode used (actually this value and number of lines in frame determines max sensor mode framerate possible), while total frame time should absolutely match video fps, otherwise, frames are sometimes dropped and duplicated, as could be observed in stock firmware, where sensor outputs frames with speed 29.97/59.94fps, while video is recorded with 30/60fps. So, only some timing values fit into these constraints, and no much options here.
 
OP
H

hc_1982

Member
Joined
Aug 19, 2018
Messages
95
Reaction score
63
Country
Ukraine
Hello @user777
please check out new test mod:

Sensor modes for 1080p30/1440p/1080p25 have been changed and timings updated. Please test with both EIS Off and On.
 

user777

Member
Joined
Jan 29, 2019
Messages
44
Reaction score
6
Location
Finland
Country
Finland
Dash Cam
GitUp Git2P
Hey @hc_1982, thanks so much for the new test FW. I will try to test as soon as possible, but my time is terribly limited during the weekdays and in the weekends the weather is typically Finnish (=crap) to do any riding.

Anyway, just this afternoon we got some sun so I took the bike out for a few tests.
I tested the G3 vs the Git2p on the mount I used last summer and with stock FWs.

Here are my first observations:
1. The stock FW is not much better in the jello effect.
2. On both cameras the 1440p is MUCH better than the 1080p in terms of details, sharpness and overall IQ.
3. The G2p is still MUCH better than the G3 in the jello effect. With the G2p I hardly saw any jello in the videos, but with the G3 in the 1080p30 and the 1440p the jello was really terrible.
4. On both cameras the best setting for fighting the jello issue was 1080p 60. I am guessing that with 60 fps the camera exposes much faster therefore reduces the jello.

I now think that the secret to G2p's great performance over the G3 is its camera bracket that maybe helps reduce the vibrations. The G3 doesn't have such a mount so it may be an issue here...
47190
 
OP
H

hc_1982

Member
Joined
Aug 19, 2018
Messages
95
Reaction score
63
Country
Ukraine
Hi @user777

Git2p with stock fw for 1080p30 is using the same sensor mode as for 1080p60 with the same timings, so jello, I think, should be no much difference between 1080p30 and 1080p60 here. This sensor mode is quite fast, as it outputs only half of full sensor resolution and with 16:9 cropping, but details are not good here, and digital zoom is quite useless higher than x1.2 zoom factor. While full readout mode (with 16:9 cropping) is used for 1440p, and it produces very detailed image, but, unfortunately it is ~2 times slower, so jello will be much higher here. The same sensor mode is used in modded fw for 1080p30 with similar results in terms of details and jello.

Git3 with stock fw is using full readout mode for 1440p (4:3), which produces 35fps at max, so it is almost ~2 times slower compared with git2p 1080p30, but it probably should be better with 1080p30, as another sensor mode is used here, which is using 16:9 cropping and a bit faster, and another mode is used for 1080p60, which is capable of producing 60fps with full readout with some cropping, and should be equal to git2p's 1080p30/60 in terms of jello, but with better details. I have switched 1080p30/50/25, 1440p30/25 to this mode in the latest test mode, so jello effect should be mostly similar here for these resolution, with some small difference between them due to timings used. The least amount of jello should be still in 1080p60 (~16.6ms per frame), then 1080p50/25 (~17.5ms per frame), and worst in 1080p30/1440p (~18.2ms per frame). As you can see, difference between 1080p30 and 1080p60 should be not very noticeable now, as it less than ~9% now, while previous version of mod had different timings for 30fps video resolutions, so, readout was almost 2 times slower. Also, I have tried to recalculate parameters for EIS for these sensor modes, so, according to my test, it seems to be better at compensating rolling shutter with current settings.

BTW, have you tested with EIS On or Off?

There are no much difference in exposure settings in bright light between 30 and 60 fps, as quite short exposures are used in both cases.
 

user777

Member
Joined
Jan 29, 2019
Messages
44
Reaction score
6
Location
Finland
Country
Finland
Dash Cam
GitUp Git2P
Hi hc_1982, just a very quick note regarding the lastest test FW. I tried it this afternoon and unfortunately the video is still suffering greatly from the jello problem.

I did tests without EIS and the videos were simply unwatchable due to shake and jello. WIth EIS on the G3 is working overtime to correct the vibrations but still...
I know that my bracket is the biggest issue here; there is just too much vibration. However, I don't have another mounting point on the motorcycle right now, and I have run out of time; tomorrow I am leaving for my long journey so I have no choice but to go back to the original FW which had the least amount of jello.

It is a shame that we couldn't solve the problem because your FW is a beautiful piece of work; it makes the video quality SO much better in every respect. But it also makes the G3 much more susceptible to jello, so for me it is a deal breaker since I need usable videos.

Let me know if you can run your own tests and we'll compare notes.
 
Thread starter Similar threads Forum Replies Date
S Git3 0
BCHobbyist Git3 22
Linhos Git3 10

Similar threads


Top