GIT2P 90 firmware mod

[naturelover]

Thank you for the thorough reply.

 

[user777]

I believe, stabilizer will work almost in the same way, but distortion correction for 170 degree lens won't be applied to image, so it may cause some "warp" in the corners of stabilized video. I'm going to add these settings for 170 FOV camera when next version of mod is released.
BTW, which video resolutions and functions are you using most with this camera?

Can't wait for the modded 170 FOV firmware!
It will make an excellent camera even better.

As for the video and functions I'm mostly using 1440p at the moment and all other functions at normal/standard modes. High bitrate.
If I want to use it on board the bike I'm using 1080p at 60 fps for the buttery smoothness it provides.

 

[hc_1982]

As for the video and functions I'm mostly using 1440p at the moment and all other functions at normal/standard modes. High bitrate.
If I want to use it on board the bike I'm using 1080p at 60 fps for the buttery smoothness it provides.

I would suggest enabling effect Black&White for video (that is just original name of effect), it will prevent clipping of shadows and highlights during video playback in media player or youtube. It "squeezes" range of pixel values to fit into TV standard, where legal range of data is 16-240, and full range (0-255), which is used by default for video in this camera, is displayed incorrectly by media players, which are ignoring and clip all these out of range values, making contrast too high. No need for this effect for .JPG images, as full range is used as standard one, or in case video will be edited and editor allows to select input data format manually.
Sensor data for 1080p60 currently is bit out of sync with video encoder, so it skips and drops frames periodically. So, if frequency in your country is 50Hz, I would suggest trying 1080p50 (original name in menu "1080p 4:3"), as it has same smoothness as 1080p60 if viewed on display with 50Hz refresh rate, correct timings, so no frames are dropped, and longer exposure possible, so image is brighter at low light, but this could be overridden by setting same shutter speed as in 1080p60 to reduce motion blur (Shutter speed 1/60s, ISO 50 (workaround to enable ISO AUTO with custom shutter speed for video)).

 

[hc_1982]

Hey guys! Finally I have found solution for frame sync issue in 1080p60, other sensor settings have been updated previously to match video recording frame rate, but I was not able to find proper timings for 60fps video mode, which originally was configured for 59.94fps, while video recorded with 60fps, so it causes frame doubling and 2-3 frames latter frame skipping every ~15 seconds of recorded video. Previous version of mod had settings close enough to correct ones, but not precise, so it worked better for several minutes, but at some point could cause a lot of dropped and duplicated frames. So, finally I was able to change recorded video fps from 60 to 59.94, which is actually used by most of the other cameras and this change should eliminate sync issues for 1080p60.
Another changes are related to LUT calculation used for color correction, and, hopefully, should make colors more natural and produce a bit smoother and cleaner image.
Also, noise reduction has been increased to reduce noise in low light with high ISO settings and autoexposure parameters have been adjusted to reduce highlights overblown in daylight and make image brighter in low light.
Experimental support for git2p 170 has been added in this mod release. Basically, it is just the same firmware, but settings for EIS (gyro) has been changed to ones used in original firmware for 170 FOV version.
Latest version of mod could be found in top post in this thread.

 

[user777]

Great work once again!
Question: how easy is it to edit the firmware? I'd like to play with increasing the bitrate in other GitUp models like the G3 and the F1.
Also when I have some free time (and better weather) I'm planning to do a video quality comparison between your modded firmware and the original firmware of the Git2p; judging on bitrate alone the difference should be significant.

On the same topic, just the other day I was looking at the bitrate of various modes of the GoPro Hero 5; from what I can see it is around 60-65 mbps.
After testing the Git2p recording sample videos, on the original firmware it does 25 - 36.5 mbps, while the modded one does 52.4 - 70 mbps, with most modes hovering around 67 mbps.
I know that bitrate is far from everything when it comes to video quality, as there's also the lens quality, image processor, etc, but this is nonetheless interesting.
Any takers to do some back to back testing with a Hero 5?

 

[hc_1982]

Changing video bitrate is quite simple for git2/3 (and other nt96660 based devices), as special tools exist for it. But there is some limit due to memory constraints, and original firmware sets bitrate for high quality quite close to it. In case value higher than this limit is set, recording won't start and camera reboot will be required. I have found the solution to overcome this limitation and used in modded firmware, which actually doubles bitrate at the moment it is passed to codec, while buffers are still allocated according real bitrate set. Modded firmware allows to set a bit higher bitrates for some resolutions (e.g. I have tested 1080p60 with 100mbps), but it seems to be not very stable in some cases. Also, there is another limitation in buffer size allocated for single compressed frame - 1Mb in original firmware, and several mb in modded one (cannot remember actual value at the moment, somewhere between 3-6Mb). It could be critical on rapid scene change, as in case of static scene, qp is set to low value, as no much difference between frames and almost no need to compress it. Then suddenly something happens (light turns on, fast movement etc), qp is still low, as it takes some time to adjust it accordingly, and bitrate jumps to 120-200mbps, and then falls down to very low values to achieve average bitrate set. In such case, compressed frame may not fit into buffer for it, the higher video resolution, the more chances for it, and video recording stops, and starts again automatically. Increasing buffer for compressed frame allowed to make it more stable with higher bitrates, but such issue may occur still under some conditions. Also, hardware is not always able to handle stream generated during such bursts.
Values of bitrates you have found by experimenting, seems to correspond closely to actual settings. I have set 67mbps almost for every resolution with high bitrate, but for 25fps, it calculates average bitrate not accurately for some reason, and it is a bit higher.
I guess, there is modded firmware for F1, but it is based on different platform (amba, same as one used in gopro) which has high bitrates. Also, currently I'm mostly experimenting with G3, which has firmware pretty close to git2p, and moving most essentials changes from git2p mod to g3 has been relatively easy, so mod for it is also available now.

 

[hc_1982]

I'm planning to do a video quality comparison between your modded firmware and the original firmware

That would be very interesting! I can hardly remember how my git2p recorded video with original firmware, and it would be nice to see if it has any improvements or what is worse now.

 

[Psychosystema]

Hello hc_1982, I've been testing the latest version of the mod and I see that the field of view doesn't change while gyro is on or not. My settings are 1080p 60fps, 1/60 shutter speed, ISO at 50.

 

[hc_1982]

Hello Psychosystema, ISO 50 actually works as ISO 3200 in latest mod (it was described in readme file), and 1/60s uses the same settings for AE for 1080p60, as normally used for 60fps. So, technically, it is equal to ISO Auto, Shutter Speed Auto for this case. Such settings would be useful for 30fps modes (1440p, 1080p30 etc), but not for 1080p60. In case you want to limit ISO for best quality, I would suggest setting ISO to 100, and no need to set shutter speed option for 60fps, as it will always be equal or less than 1/100s (for Frequency 50Hz).
Does EIS works at all for you with this mod?

 

[Psychosystema]

Hello hc_1982, thank your for your advice. EIS doesn't seem to work with this mod for me. Is it broken?

 

[hc_1982]

Yes, I'm testing it right now, and it looks like it does not work with 1080p60, while seems to be ok for other video resolutions. Probably, it is related to framerate changes from 60fps to 59.94fps in latest mod. I will try to investigate this issue and fix it.

 

[hc_1982]

New version of mod with EIS fixed for 1080p60 and several other updates is available.
@Psychosystema, please check top post of this thread for download and more details.

 

[Psychosystema]

Thank you so much for your quick fix! EIS is working again with 1080p60.

 

[user777]

@hc_1982 I just returned from a few extensive tests of your latest firmware on a Git2p. Is it possible that your firmware could introduce rolling shutter issues (jello effect) to videos? I took some 2K videos and there is some VERY noticeable jello effect on the trees & details on either sides of the road.

I used this same camera extensively last summer with stock firmware on the same motorcycle and same mount and I don't remember seeing this issue at all.

 

[hc_1982]

@user777, I guess rolling shutter should be the same for 1440p on stock and modded firmware. Probably, it will be higher for 1080p30 on mod, as sensor mode used for it is shared with 1440p, while stock firmware is using mode 1080p60 for it, which has pixel binning and reads out faster, as result less rolling shutter, but with worse details and "stairs" artifacts. Another possible reason for higher rolling shutter issue is sensor timing used for full sensor readout mode, I will check if it possible to improve it with different settings.
Which video resolution has been used last summer with stock firmware (1440p, 1080p30 or 1080p60)?

 

[user777]

Maybe I didn't explain correctly. Today I didn't use the 1440p resolution but the highest available. The video information shows it to be 2880x1620, 30fps.
Last summer I only used 1080p30.
All videos have had EIS on.

 

[hc_1982]

I think, this highest resolution (2880x1620) is good for static scenes only, as its bitrate is too low and probably different codec parameters are used due to hardware limitations. 1440p seems to be noticeably better in motion and with less jello, as for me. Although the same sensor mode is used in both cases, so rolling shutter will be the same. 1080p60 or 1080p50 should be used with this mod (1080p50 is preferred for 50Hz Frequency, especially in low light, but it might look not so smooth, if your display is set to 60Hz refresh rate), in order to get the same rolling shutter level as you have experienced on stock 1080p30.

 

[user777]

@hc_1982 thank you for yourgreat efforts and suggestions! I will try again with 1080 video and 1440p just to see if it changes anything. In any case I've also gotten an anti vibration mount which I will try soon to see if it makes any difference.
If the jello effect stays the same I'll have to go back to stock.
On a different note, as promised I have done some interesting photos & video comparisons between the stock firmware and the modded one so I'll come back to this soon.

 

[user777]

As mentioned earlier, I made some back to back tests between the stock FW and the modded FW of our friend @hc_1982.
Regarding the photos, there is *absolutely* no comparison between the two firmwares; the modded IQ is so far better that it's almost a joke to even put them side by side.

Simply put, on the stock firmware the Git2p applies far too much noise reduction and very aggressive sharpening to the point that photos almost become cartoon drawings. They are perfectly fine viewed at web resolutions, but blown up full size on a decent screen the stock images leave A LOT to be desired.
Here you can see an example, magnified at 100%. On the left is the stock FW, on the right is the modded.
Stock settings standard / Modded settings standard except for sharpening which is set to high.
Correction: Optimum IQ settings for modded FW were: Sharpness HIGH, Color COLORFUL, EV -1/3



As I said; simply no comparison! The right hand side image looks so much better in every way that it could well have been taken with a different camera altogether. The modded FW seems to apply very little noise reduction, with just a tad of chromatic noise creeping in but it's far preferable to the overcooked mess on the left. The only issue that I had (and it was the same for the video side) was that the modded FW tended to easily blow out the highlights. Especially clouds in the sky would quickly become white blobs, so I guess the curves have been tweaked a bit. I solved the issue nicely by setting the EV to -1/3.

Now, what I do like about the stock image is it's color and tonal balance. I think that the modded FW image is a bit cold and low contras for my taste. A few Photoshop clicks later and we get this:


Honestly, I couldn't be happier about such result. It easily beats most compact P&S cameras and I'd be more than happy to put this up against the best snapshots from even the latest and greatest GoPros.
Now, about the video IQ, well... this is were it gets a bit tricky.

 

[user777]

To begin with let me say that contrary to the photos, I do like the video IQ of the stock FW. Yes, viewed at 100% it shares a lot of the bad qualities of the photos (oversharpening, truckloads of noise reduction, etc) but on the other hand a video is not meant to be scrutinised like a still image.
When played back you get a lovely tonal character, nicely saturated colours, smooth motion with very little/no jello effects... Yes the bitrate is rather low, yes you can see the noise reduction smothering details on trees and road, but on the whole the videos are very nice indeed.

Now, there is no question if the modded FW's videos are better or not. They are.
But they are also far less "impressive" than the stock ones at least on stock settings.

Using a stock clip as a guide I tried many settings to get as close to the original "punchiness" as I could. I took about 20 clips and a fair amount of side by side viewing but here's my issues with the modded FW video image quality:

1. Most serious issue: pronounced rolling shutter (jello effect). On the test clips I hadn't seen anything as there were no vibrations, but on the motorcycle two days ago I noticed some serious jello effects especially around the edges of the video (trees, power lines, hills, sky etc) On the stock FW I had no issue but all those clips were done at 1080p30, so it may be just wrong settings. I'll run some tests again and report. If this doesn't go away though it is potentially a deal breaker.

2. Second serious issue: highlight clipping. One thing that was immediately obvious on the modded videos was that cloud details were mostly blown. The stock video kept all the cloud details nicely preferring to underexpose the whole scene a bit rather than lose the details. On the modded videos, the scene was slightly brighter but you got big blank/white patches on the sky where grey clouds should be. Maybe @hc_1982 modified the curves a bit resulting in this highlight clipping?

2. Rather soft footage. Normal sharpness is a bit low IMHO. High sharpness solves the issue nicely but could go even a bit further.

3. Low color saturation and contrast on stock settings. While the color saturation can be nicely fixed with the "Colorful" setting, the lowish contrast is not so easily changed. The high contrast setting is simply too aggressive. Of course this is a moot point as everything can be tweaked in post.

Eventually I settled on the following settings for the best look out of the box:
- Sharpness: HIGH
- Contrast: LOW
- Colours: COLORFUL
- Bitrate: High

Below you can see a side by side comparison of the stock clip (Left) vs the best mod clip (Right).
Overall IQ:


Immediately apparent that the modded clip has a better tonal character and equally pleasing colours. Again the eagerness to blow the highlights (look at the sky on the right) but this can be fixed by setting EV to -1/3. I didn't set it as I felt it wasn't such a huge issue here, but YMMV.

Details @ 100%:





Details on the modded clip are visibly better, especially on the trees and fine details, but I think the difference is not as pronounced as with the photos. I guess we are bumping up the limits of what this little camera can do?

All in all the modded FW's videos seem to come out very nice indeed, BUT I reserve final judgement due to the jello issues.

Full clips for you to download.
Stock FW clip. All settings standard.

UPLOAD.EE - File does not exist

File does not exist. Upload.ee

www.upload.ee

Modded FW clip. Best settings found as above.

UPLOAD.EE - File does not exist

File does not exist. Upload.ee

www.upload.ee