Author Topic: Camera - is it IMX586?  (Read 3727 times)

sup

  • Newbie
  • *
  • Posts: 35
    • View Profile
Camera - is it IMX586?
« on: October 14, 2020, 05:24:30 pm »
Planet announced the camera for Astro a while back to be 48MP from Sony. To me it sounds like the IMX586 model, so once again the true resolution is more like 12MP than 48 MP (https://www.nextpit.com/48-mp-or-not-look-at-sony-new-camera-sensor), but still a nice improvement over the Cosmo. Apparently, the final results can be quite different (and that is from phones that use an additional camera): https://www.notebookcheck.net/Sony-IMX586-Comparison-Review-Five-48-MP-smartphones-face-off-in-a-camera-duel.428434.0.html.

Do you think this is indeed the camera Astro will come with? And is there any hope it shoots decent pictures with the default settings, or will we need to wait for some hack like the lovely Shuntcap's GCam for Cosmo? https://www.oesf.org/forum/index.php?topic=35912?

Daniel W

  • Sr. Member
  • ****
  • Posts: 372
    • View Profile
Re: Camera - is it IMX586?
« Reply #1 on: October 15, 2020, 02:46:39 pm »
Based on this Wikipedia page, and some web searches to verify, it might be either Sony IMX582, IMX586, IMX598 or IXM689, all of which are 48 mega-photosite sensors - with Quad Bayer filters, presumably for better noise reduction and dynamic range.

According to this Sony press release, the IMX586 can, in bright light (and with proper software), spatially resolve 48 million pixels (though with a bit blurrier color than with a classic Bayer filter). Reasonably, that's true for the other Sony sensors too.

That said, the 0.36" type Samsung S5K2X7 sensor in the Cosmo, should, on the hardware level, be able to spatially resolve 24 million pixels (lest Samsung is outright lying), though neither the stock MediaTek software nor shuntcaps adapted GCam, can (currently, anyway) make it output such images.

I think it's reasonable to regard the Cosmo sensor as optimized for 6 megapixel images and, by extension, the Sony sensors as optimized for 12 megapixel images. In my opinion, that's about right for a phone, as it allows 4K filming for those so inclined, while not creating unreasonably large files when taking photos with the default settings.

For special occasions, though, it would be nice if we could force the Astro camera to 48 megapixels. I am a bit afraid though, that the stock drivers will snap 12MP pictures and interpolate... :P

sup

  • Newbie
  • *
  • Posts: 35
    • View Profile
Re: Camera - is it IMX586?
« Reply #2 on: October 29, 2020, 02:13:42 pm »
Yes, I am also afraid so. Not sure how to communicate to Planet computers that they should deliver better software for the camera, they seem to be clueless about it :-(.

sup

  • Newbie
  • *
  • Posts: 35
    • View Profile
Re: Camera - is it IMX586?
« Reply #3 on: October 29, 2020, 02:14:16 pm »
I wonde if Shuntcap ordered an Astro. Because without him, we might be lost.

shuntcap

  • Jr. Member
  • **
  • Posts: 84
    • View Profile
Re: Camera - is it IMX586?
« Reply #4 on: January 16, 2021, 06:13:28 am »
I took the liberty of transcribing Planet's answer to a backer's question about the Astro's expected camera quality.  This is from Indiegogo's Astro Slide update #20, January 15, 2021:

"The second video addresses the questions that we were asked by you, our backers, this week."

https://youtu.be/SgQ8NTituUA 8:02-9:24

CHRIS BIGNELL - XL COMMUNICATIONS: "Entienne asks about the camera quality in terms of uh, you know, the, within the Cosmo, the MediaTek stack did not give um, did not support the full capabilities of the sensor, uh, that I know is something that has been improved over time, uh, is there likely to be, uh is that likely to be a problem with Astro Slide, or is that something that we think we've resolved?"

DR JANKO MRSIC-FLOGEL - CEO, PLANET: "So we think there'll be a huge improvement in the camera, uh, one because uh, the uh, the sensor is a Sony sensor, it's a high quality, you know, Tier 1 sensor, and uh, uh, the pre-- it's 48 megapixel and high quality with uh, um, standard oriented, so basically uh uh um uh, a square-type orientation, um, um, a square-type organization of the CMOS, whereas the uh, CMOS sensor, whereas the, uh, the previous sensor was a 24, uh, megapixel sensor from Samsung which has a hexagonal organization of CMOS, so basically it's, it's far be-- the n-- the new sensor not just that it has a larger resolution but it's uh, uh, the-- the organization of pre-- the CMOS sensor is actually better so you'll get a, a far superior performance."

CHRIS BIGNELL: "Good to hear, thank you, and I'd also like to thank Felagund and Colin W., two other backers that asked the same question around the camera."

God bless them for trying and I wish them the best, but wow... Planet has absolutely zero understanding of the Cosmo camera issue. The backer understood.  He obviously read our Cosmo camera forum posts.  Planet did not!  The issue lies entirely with the MediaTek software stack!  It's not a question of needing higher resolution.  It's not a question of "CMOS hexagonal organization" versus "CMOS square organization" (whatever that means).  The sensor itself, as we have proven, takes sharp, detailed, beautiful photos with the right software: Google Camera ported to the Cosmo (https://www.oesf.org/forum/index.php?topic=35912.msg297145#msg297145) which completely bypasses the MediaTek imaging stack for photos.

Regarding hexagonal or square pixel arrays, Planet's explanation truly left me scratching my head.  Samsung's Tetracell and Sony's Quad Bayer are both Bayer filter mosaics in a square pixel array, albeit with multiple adjacent pixels for each color.  But the colors are still staggered.  Neither is a hexagonal pixel array, which is very uncommon, and if the Cosmo's Samsung sensor were, it would actually be a computational improvement (but not an image improvement).  And neither has any bearing on MediaTek's inability to properly process raw image data.

It's also amazing how the Cosmo camera issue magically "has been improved over time."  Wonder how that happened.

I think GCam would produce fantastic photos with a 48MP Quad Bayer sensor.  But for the record, I am not backing the Astro.

shuntcap

  • Jr. Member
  • **
  • Posts: 84
    • View Profile
Re: Camera - is it IMX586?
« Reply #5 on: January 16, 2021, 07:15:31 am »
Planet announced the camera for Astro a while back to be 48MP from Sony. To me it sounds like the IMX586 model, so once again the true resolution is more like 12MP than 48 MP (https://www.nextpit.com/48-mp-or-not-look-at-sony-new-camera-sensor), but still a nice improvement over the Cosmo. Apparently, the final results can be quite different (and that is from phones that use an additional camera): https://www.notebookcheck.net/Sony-IMX586-Comparison-Review-Five-48-MP-smartphones-face-off-in-a-camera-duel.428434.0.html.
None of the phones in that review uses a MediaTek chipset.  They are as follows:
Asus ZenFone 6: Qualcomm                                                   
Honor View 20: HUAWEI                                                     
OnePlus 7: Qualcomm 
Xiaomi Mi 9: Qualcomm
ZTE Axon 10 Pro: Qualcomm

To see what a 48MP sensor looks like in a MediaTek phone, take a look at GSMArena's photo comparison tool:

https://www.gsmarena.com/piccmp.php3?idType=4&idPhone2=8310&idCamera1=300738&idCamera3=300610

In this link I have selected three phones:
Oppo Reno4 Pro 5G, 48MP (MediaTek chipset)
Motorola Moto Z Play, 16MP (standard sensor, not Quad Bayer or Tetracell)
OnePlus 7(48MP, Qualcomm chipset)

I chose the Oppo Reno4 Pro 5G because I figured it's probably the newest 48MP MediaTek-based phone in the GSM Arena database.

I chose the 2016-era Moto Z Play, with its "smaller" 16MP sensor, as a benchmark.  I also happen to own one because I backed the failed Moto Keyboard mod in 2017.

I randomly chose the OnePlus 7 from the list of phones in the article that sup provided.

In good lighting, the Moto Z Play wins easily in texture detail and resolving power.  Look at the street map: well-defined edges and small but legible text.  Look at the fabric patches: sharp, detailed texture.  It trails the OnePlus 7 in the paper currency detail, however, but that's the exception.

Even in low light (click the "Low light" image), the Moto Z at worst matches the image quality of the Oppo (and smokes the OnePlus).  At best, it looks much more natural (to me) than the Oppo with its interpolated image.  I would have expected the Oppo's 48MP sensor to perform much better than the Moto Z Play's in low light, but this is apparently not the case.  The Moto Z Play also easily beat the ZenFone 6 (false color artifacts and noise), the Honor View 20 (smeared away all detail), and the Xiaomi Mi 9 (smeared away all detail).  The ZTE Axon 10 Pro is not available in the comparison tool.

My point of this demonstration is two-fold.  First, according to the marketing, the 48MP sensors should be blowing away the 16MP in detail resolution and texture reproduction, especially in low light, but in almost all cases they fall behind, often way behind.  Second, the Oppo (MediaTek) images looks pretty bad in my opinion, with lots of jagged edges and halos due to interpolation and oversharpening.  And that's an improvement over the first Oppo 48MP model I picked, the Oppo F11 Pro.  Compare the route colors on the map with the Moto Z Play or the OnePlus 7.  They're all wrong, just plain wrong!  They look like the burnt colors from the Cosmo under the MediaTek stack.  So there's a smidgen of hope that the Astro's color reproduction will be improved over the Cosmo's using MediaTek-based image capture, but don't expect much of anything for sharpness and texture.  And above all, don't fall for the marketing hype.