Mobius Grainy at Night?

[Dashmellow]

While we're on the subject, I thought it might be helpful or perhaps interesting to at least some of us here to explain a little further and illustrate what is happening in the above video example of the Red - Cyan color shift. The Mobius DSP is attempting to map out the RGB (red-green-blue) color values it sees and balance them towards neutral.....ie: "white balance". For some reason, in certain early firmware versions, the algorithm (sometimes) overcompensates the balancing process and overshoots the midpoint. In this instance, it is seeing some red suddenly coming into the frame and it calculates that it needs to add cyan to compensate for the red but it adds way too much and for too long a period of time. Cyan (blue-green) is the complementary of Red on the RGB color wheel and is on the exact opposite side of the color wheel circle. White is in the middle and would be the balance point. (again, this is why it is called white balance) So, in the video example we see the whole scene suddenly shift all the way to a cyan color cast when the red object comes into view and it stays that way for a while. Personally, I've experienced this cyan color shift along with the also well documented yellow cast shifting problem when a large blue object enters the frame. Of course, as can be seen on the color wheel, blue and yellow are complementary opposites.

Here is what it looks like on the RGB color wheel:

 

[russ331]

I myself have not had this with 0.59fw as yet. Maybe it is rare? It was the orininal FW on my mobius that I have had since release over a year agos.

Firmware v0.59 was released at the end of last November.

 

[Dashmellow]

To quite frank be I feel unwelcome heres and any speach of free seems to be frowned onto. You all seems to act like we attack your loved mobius and defend without open minds.

With rare exceptions DCT has been a friendly and supportive place with a very positive atmosphere and personally have found this to be more so here than other forums I have participated in. Rickardo, your opinions and experiences are welcome here just like anyone else. The fact that others may have a different opinion or may disagree with you should not be a reason to believe that you have no right to free speech or that your opinions are any less valid. As long as we are respectful and civil with one another we can share opposing opinions and perceptions, offer evidence to support them and still do so in a friendly way. Discussions where people argue their point of view can have value. That is in a small way part of the value and enjoyment of our interaction here, no? I have certainly learned a lot from others here on the board whose opinions I don't always share as my own. Nevertheless, Rickardo, I am sorry you have felt unwelcome here and I would like to apologize for any part I may have played in making you feel that way.

 

[Odi]

I just bought 4 Mobius with B Lens from Banggood a couple of weeks ago to install as my first dash cams (front+rear) in 2 cars, and before I install them (and while waiting for the hardwire kits from Asia), I figured I would run some comparison tests between the firmwares. This is the closest I could mount them together, the videos are lower than where I will finally mount because I used 2 suction smartphone clips to hold the metal bar.

They are mounted with oldest firmware on top and newest on the bottom. This makes the newest on the left and oldest on the right when mounted in the car.

These are the relevant settings, all set the same
Mobius B Lens
60Hz Artificial Light
Wide Dynamic Range On
Video Data Rate High
Video Clip Length 3 minutes
Video Frame Rate 30fps
Video Resolution 1920x1080
Field of View Wide
Rotate Video 180 On

Both videos have the following layout:
0.59: upper left
1.13: upper right
1.17: lower left
1.20: lower right

Night comparison (be sure to change to 1080p)

The first few minutes of the video shows the advantage of 0.59, so I posted up the raw videos as well. If you want other segments of the video raw, just let me know and I can upload those too. Step 2 firmware cannot make out a lot of the peripheral trees, while Step 1 seems to show the trees quite well. 32 seconds in, you can see a lot of trees that are just empty darkness in Step 2 firmware. 1:40 - 1:45 also show the trees in Step 1. 3:25 shows the color shift to orange, so Step 2 seems to reproduce colors much better. This can also be seen in the Dusk comparison at 8:52 with all the yellow signs and yellow paint on the road.

Re-encoded at 8000kbps split screen video
0.59 raw
1.13 raw
1.17 raw
1.20 raw

Dusk comparison (be sure to change to 1080p)

It took me a week to find the time to figure out how to put the videos together into one video, but any subsequent tests should reach the internet much faster. I have currently reloaded the firmware to

0.59 -> 0.47
1.13 -> 0.53
1.17 -> 0.57
1.20 -> 0.59

and will perform another test soon. After that, my plan is to pick the best Step 1 among these 4, and compare it with 1.20 WDR On, 1.20 WDR Low Light, and 2.1X WDR Low Light.

I used AviSynth + avs2avi + x264vfw to put the 4 videos into 1 and ran 2 pass encoding at 8000kbit/s to make the file size reasonable to upload yet with good quality. My initial attempt using VSDC Video Editor resulted in ridiculous file sizes (16 min video, source 2.1GB each from RegistratorViewer with 1020ms overlap, ended with 8GB file even though the resolution stayed the same -- each source video shrunk to 1/4th original size -- plus the entire video only had 1 keyframe)

RegistratorViewer with 1020ms overlap to combine 3 min files into one file
Avidemux to trim the front and back of the video at keyframes

Script I used for AviSynth to shrink and align the videos to the same frame

h1=AviSource("night059.mov",audio=false,fourCC="X264").BilinearResize(960,540)
b1=BlankClip(clip=h1,length=3,color=$000000)
c1=UnalignedSplice(b1,h1)
h2=AviSource("night113.mov",audio=false,fourCC="X264").BilinearResize(960,540)
h3=AviSource("night117.mov",audio=false,fourCC="X264").BilinearResize(960,540)
b3=BlankClip(clip=h3,length=5,color=$000000)
c3=UnalignedSplice(b3,h3)
h4=AviSource("night120.mov",audio=false,fourCC="X264").BilinearResize(960,540)
b4=BlankClip(clip=h4,length=10,color=$000000)
c4=UnalignedSplice(b4,h4)
l1=StackHorizontal(c1,h2)
l2=StackHorizontal(c3,c4)
StackVertical(l1,l2)

Then I just ran avs2avi twice, once with Multipass 1st pass, and once with Multipass nth pass.

 

[reverend]

There's absolutely no arguing with that - amazing work thank-you

Possibly a daft question but if you run them all on the same firmware with the same setting do they all produce the same video or are there differences there as well? Just wondering if they naturally have slight differences anyway.

 

[Dashmellow]

There's absolutely no arguing with that - amazing work thank-you

Possibly a daft question but if you run them all on the same firmware with the same setting do they all produce the same video or are there differences there as well? Just wondering if they naturally have slight differences anyway.

Very good question! In the discussions of graininess and color shifts, I've wondered if individual camera quirks and manufacturing differences or tolerances might possibly contribute to people's differing experiences with them.

 

[reverend]

That's exactly where I'm coming from mate, did wonder if there are tolerances between individual units which could affect stuff like this in addition to the firmware.

I've already noticed other cams show different results to what I get - my G90 for example isn't as sharp as some seem to be nor are my GT680s. The A118 and LS460W are razor sharp though - read somewhere that the A118 could be focussed in a jig instead of by human hand which is why they're all bang on. That had to have been one of Rick's posts.

 

[jokiin]

Just to clarify, if the CMOS is not on the mainboard then they can be setup in a jig, the focus is still done by a person though so human error is still possible

Even when onboard the camera may still be placed in a specific mount for the purpose of setting the focus, some factories do this well and get consistent results, some don't

 

[Odi]

I'll try that too, I had the thought in the back of my mind, which is why I kept the top one as oldest, instead of keeping it at 0.59 and making the bottom one 0.47 (saving one update in the process).

 

[russ331]

I just bought 4 Mobius with B Lens from Banggood a couple of weeks ago to install as my first dash cams (front+rear) in 2 cars, and before I install them (and while waiting for the hardwire kits from Asia), I figured I would run some comparison tests between the firmwares...

Thanks very much for this. I see it's only your second post at DCT, yet so accomplished & useful. I look forward to more of your comparison videos, etc.

 

[Dashmellow]

@Odi, I agree with Russ. This is an excellent, well crafted and very objective comparison that tells the story well and I look forward to your next round.
Actually, there's enough room on that piece of angle iron so I think you really need to add four more Mobius cameras so you can compare everything at once! (Just kidding......)

 

[Signhere]

Are you proving no color shift as you have a truck also red building on video? Or shift and I not see?

What bit you refer please? Nice clear video.

I like speedo you have in dash.

Not a proof - just to show an example of no color shift under similar conditions.

It's a GPS "speedo" (very fast and accurate). Can also be used as a bluetooth GPS receiver ( I got it six years ag0 ! ).


18 mbps w/ WDR off ...

 

[Odi]

Here is a shot of them mounted in the car

The oldest firmware sits at the left of the picture (right side of car), and the newest sits at the right of the picture (left side of car). So if you assign numbers from 1 - 4, with 1 being the leftmost in the picture

1: 0.59 / 0.47 / upper left
2: 1.13 / 0.53 / upper right
3: 1.17 / 0.57 / lower left
4: 1.20 / 0.59 / lower right

These are the relevant settings, all set the same. The bold shows changes from the last test I did (basically set Standard instead of High)
Mobius B Lens (except 0.47 which cannot set it)
60Hz Artificial Light (except 0.47 which cannot set it)
Wide Dynamic Range On
Video Data Rate Standard
Video Clip Length 3 minutes
Video Frame Rate 30fps
Video Resolution 1920x1080
Field of View Wide
Rotate Video 180 On

Video has the following layout:
0.47: upper left
0.53: upper right
0.57: lower left
0.59: lower right

I cut the path to the first 6 minutes of the first night video to save time driving and encoding. When watching the first night video, I felt that the last half didn't really contribute much, as it was all very well lit streets (and I can't seem to find streets in my area with no street lights...it's something I never paid attention to until I was trying to make videos like these and realized there are no nearby car parks, tunnels, or dark streets).

Step 1 comparison (be sure to change to 1080p)

I think 3:40 illustrates how much more advantageous 0.59 is compared to the others. The other Step 1 firmwares were much worse than I expected that I actually went into mSetup to double check that I had all the options correct after reviewing the videos. Compare it to 3:29 of the first night video I posted, and it seems that whatever fix was in 0.59 was maintained in Step 2, except that 0.59 is marginally more readable. Throughout this new video, 0.59 sees a lot more near my headlight area of illumination than the other Step 1 firmwares.

0.59 hardware comparison (be sure to change to 1080p)

The videos show that there are slight angle differences based on how much dash is showing, but it didn't seem to affect the quality. I think the camera hardware is sufficiently identical that the variances in screw/hex nut position made more difference in the videos. So this baseline tests suggest that the other videos are mostly due to firmware settings.

When I find more time, I will try 0.59 WDR On compared to 1.17 / 1.20 / 2.10 with WDR Low Light. The reason is right now, if I were forced to choose and not do anymore tests, I would choose 0.59. But Low Light has been said to improve things, with the flashing problem as a drawback. I'm also open to any tests someone may think up. One thing I was pondering was changing the White Balance from Auto. 0.59 didn't have any WB options, so I'm wondering if there's a WB setting in Step 2 that would be more similar to 0.59. Also, I have read that exposure +1 was too much. Not sure if it's even worth barking up that tree. My hardwire kits have arrived, but I'm ok with delaying my installing of the dash cams for a bit if we can figure out a setting to make the newest perform similarly to or better than 0.59.

 

[russ331]

I'm also open to any tests someone may think up.

Great stuff. Much appreciated. I'd be interested to see if there's any practical difference in night time video clarity between the Wide & Narrow FOV settings.

Also, whether there's any noticeable difference in video quality between the Low, Standard & High data rate settings.

 

[dirkzelf]

I'm surprised also that so no one is mentioning the older firmware problem with start up that so many have posted about, especially diesel owners. Owners who had to wait at ACC before start up were having to unplug and reinsert the USB plug after the car started to get their Mobius' running. V2.10 has made startup almost flawless.

Yes, maybe v0.59 is better at night, yet new firmwares were introduced to improve so many usability issues, for example color shift, color balance, saturation, adding motion detect, start up and sd card compatibility. Not to mention so many extra user requested items like 25fps and the ability to set more settings to truly have 3 usable Modes.

The Mobius is an action/dashcam video recorder, so IMHO it needs the best possible recording.

Yes it's great to have many improvements on features, but what good are they if it's main function, good video quality, is being affected?

In other words, "what use is a nice set of teeth if you have nothing to laugh about?"

 

[dirkzelf]

When I find more time, I will try 0.59 WDR On compared to 1.17 / 1.20 / 2.10 with WDR Low Light.

Is this the LowLight WDR test you're talking about?

 

[dirkzelf]

Unlike the above post in 2.10 I don't seem to be having issues with unusable video, there just seem to be vertical lines throughout the entire frame. I have yet to see any flickering or flashing that others have described either, and night quality doesn't seem any worse, there are just the annoying bars in the dark spots.

I have no idea why yours is OK and mine isn't.....

What other settings did you use?

Standard or wide lense?
Video Data Rate?
White balance?
Artificial Lighting Frequency?
Video Rotate?

 

[lamboboy732]

I have no idea why yours is OK and mine isn't.....

What other settings did you use?

Standard or wide lense?
Video Data Rate?
White balance?
Artificial Lighting Frequency?
Video Rotate?

Wide lens
Standard rate
Rotated 180
The rest is all stock

 

[Odi]

Is this the LowLight WDR test you're talking about?

Yes, I got busy IRL and haven't had a chance to play with the Mobius since. I uploaded that, and then had a problem with the Wide vs Narrow footage. I figured out that I wanted to test 0.59 WDR On and 2.10 WDR Low Light in both Wide vs Narrow, since those were the two I liked the best. However, the 2.10 Wide ended up being very dark, like some other video here. I haven't had time to investigate if it was a loose camera ribbon, or if reverting to 0.59 would fix it.

The bad camera went from 0.59 -> 1.17 -> 2.10. The other cameras were ok: 0.59 / 0.59 -> 1.20 -> 0.59 / 0.59 -> 2.10

I was going to roll back to 0.59 first to check if it's software or hardware, but with several visitors to my house kicking me out of the extra bedroom, my now 9 month old becoming very mobile, and deadlines at work, the rig has been sitting in a closet...

 

[antcon]

where did the thread go? It's 11-30-14