LOFIC sensors are the future...

[kye]

Cined just published their tests of the DJI OSMO Pocket 4P camera, with the new LOFIC sensor, and oh boy does it deliver!

Highlights from their testing (of the wide camera):

• 16 stops at SNR=2, and 17.2 stops at SNR=1, with no internal NR (better than the Alexa 35 and any other camera they've ever tested)
• 10+ stops of latitude (second only to the Alexa 35) and potentially limited by the 10-bit codec and maybe not the sensor
• 11ms rolling shutter (not bad, and better than DGO designs)
• Their studio lights weren't powerful enough to clip the sensor, it has that much DR!

The telephoto camera from the DJI isn't a LOFIC sensor and despite testing very well it's not at the same level of performance.

The full test: https://www.cined.com/dji-osmo-pocket-4p-lab-test-rolling-shutter-dynamic-range-and-exposure-latitude/

In watching their podcast presentation they revealed a bunch of other interesting info:

• This is the first of these (tiny consumer) devices that isn't trying to adapt itself to the scene, so it's a "real camera" in the sense that it's not heavily processing things
• They mentioned they had the iPhone 17 Pro in for testing and even when it was in full manual mode, just by moving the focus point it would change the test results etc
• As the main DJI camera is ~20mm FOV, they had to develop a new testing setup so the bright patches of the Xyla didn't reflect from the camera onto the dim patches etc
• LOFIC sensors are likely to be coming into smartphones soon, as this is where the money is in the R&D
• Sony are working on a smaller version of a LOFIC sensor which would be suitable for smartphones and also for the telephoto camera on the DJI

Also, the OG BMPCC got a mention, as they were talking about how it would be great to have such a sensor in a "real" camera body and with interchangeable lenses etc, but MFT sensor size isn't that far away from this, so that could also happen too.  Panasonic went DGO and maybe would do this too..   I'm stoked for the GH8!!

Just incredible, and I second the idea of fitting one of these to a body like the OG BMPCC!

Images.

Waveform plot for the DJI OSMO Pocket 4P 20mm wide D-Log2 ISO1600

Imatest result for the 20mm wide camera, D-Log2 at ISO1600

Latitude tests of the wide camera:

The 4-stop over, which due to their lighting restrictions was still probably 1-2 stops from clipping:

 

[PPNS]

video surveillance is about to get REAL good!

[Anaconda_]

1 hour ago, PPNS said:

video surveillance is about to get REAL good!

The 'Enhance' feature on CCTV is not longer reserved for detective series.

[sanveer]

I suggested in the other group that Panasonic should put LOFIC sensors in their ILCsand someone went on about mythical organic sensor, which quite frankly may remain just that.

Panasonic should just put LOFIC sensors in all their new cameras, and ditch Sony if it doesn't improve sensor tech enough. 

I would love to see an Even Smaller GX camera with a 20-50MP 1inch LOFIC sensor. The LX10 is frankly too large for most pockets. 

[kye]

12 minutes ago, sanveer said:

I suggested in the other group that Panasonic should put LOFIC sensors in their ILCsand someone went on about mythical organic sensor, which quite frankly may remain just that.

Panasonic should just put LOFIC sensors in all their new cameras, and ditch Sony if it doesn't improve sensor tech enough. 

I would love to see an Even Smaller GX camera with a 20-50MP 1inch LOFIC sensor. The LX10 is frankly too large for most pockets. 

Apparently Sony is working on a smaller LOFIC sensor now, so swapping to LOFIC may well mean sticking with Sony, especially if you already have a deal of some kind worked out.

[ac6000cw]

6 hours ago, sanveer said:

I suggested in the other group that Panasonic should put LOFIC sensors in their ILCsand someone went on about mythical organic sensor, which quite frankly may remain just that.

Panasonic should just put LOFIC sensors in all their new cameras, and ditch Sony if it doesn't improve sensor tech enough. 

I would love to see an Even Smaller GX camera with a 20-50MP 1inch LOFIC sensor. The LX10 is frankly too large for most pockets. 

AFAIK Panasonic don't use Sony for all of their sensors - they've implied the GH6/GH7/G9ii sensor is not from Sony (I think the words were 'it's not made by who you think it is' in response to a question at a launch event).

OmniVision are also in the LOFIC game - the OV50X50 is a 50MP 1" sensor with both dual-gain and LOFIC - https://www.ovt.com/products/ov50x50/

Quote

Ultra High Dynamic Range 1‑inch Image Sensor for Movie-Grade Video Capture in Flagship Smartphones

The OV50X50 CMOS image sensor is a 50‑megapixel (MP) sensor with a 1.6‑micron (µm) pixel in a 1‑inch optical format designed for flagship smartphones that require high dynamic range (HDR) video and preview with single exposure, excellent low-light performance, fast autofocus and high frame rates.

The OV50X50 supports 4‑cell binning for 12.5MP at 180 frames per second (fps) and 60 fps with three-channel HDR. It offers premium-quality 8K video with dual analog gain (DAG) HDR and on‑sensor crop zoom. OMNIVISION’s TheiaCel® technology further expands single exposure HDR close to 110 dB—the highest range possible in smartphones. The sensor also supports quad phase detection (QPD) for best-in-class autofocus performance. The OV50X50 is built on OMNIVISION’s PureCel®Plus‑S stacked-die technology, enabling high resolution with 1.6 µm pixels.

The OV50X50 is sampling now and will be in mass production in Q3 2025.

Applications
  • Smartphones
  • Video conferencing
  
Features
  • Active array size: 8192 x 6144
  • Maximum image transfer rate:
    - 50MP (8192 x 6144): 30 fps
    - 12.5MP (4096 x 3072) linear: 180 fps
    - 12.5MP (4096 x 3072) DCG + LOFIC: 60 fps
    - 12.5MP (4096 x 3072) HCG + LOFIC/LCG: 90 fps

'TheiaCel® technology' is their version of HDR capture including LOFIC - https://www.ovt.com/technologies/theiacel-technology/

 

[MurtlandPhoto]

I’m getting a Pocket 4p in hand in the coming days. Looking forward to seeing how it matches up with the S1ii and Pyxis 12K. The videos I’ve seen truly are remarkable.

[ND64]

Big portion of that 16 stops is unrecoverable due to heavy baked in raw processing and NR. LOFIC low light performance is intrinsically poor, so they have to use two or even three readouts with conventional approach and combine them all (which is not very linear) then add some sharpness tricks to reconstruct the degraded quality. At this sensor size and consumer level, nobody cares and the only thing matters is that it makes truly HDR image. To implement this tech in bigger sensor for pro bodies, that process should be way more refined.

[Ilkka Nissila]

1 hour ago, ND64 said:

Big portion of that 16 stops is unrecoverable due to heavy baked in raw processing and NR. LOFIC low light performance is intrinsically poor, so they have to use two or even three readouts with conventional approach and combine them all (which is not very linear) then add some sharpness tricks to reconstruct the degraded quality. At this sensor size and consumer level, nobody cares and the only thing matters is that it makes truly HDR image. To implement this tech in bigger sensor for pro bodies, that process should be way more refined.

I noticed when watching various videos made with the Pocket 4P using my TV that the images had a processed appearance, especially faces seemed overly plasticky in appearance, like they were not of real people. This could be due to image processing (noise reduction & sharpening) and partly due to the characteristics of the lens. However, the dynamic range did seem spectacular and it is only this processed appearance of the main subject that bothered me. 

Is the processing something the manufacturer applies due to smaller area of the photodiode vs. conventional BSI sensor, the small overall sensor area or due to the expected requirements of the user base (which may be social media mainly viewed using small mobile device screens, and used to processed outputs from mobile device cameras)? The Pocket 4P image appearance wasn't that bad compared to typical mobile phone output, I would say it was very good in that category, but the overly smooth and dolled-up appearance of faces did bother me and it would make me question whether I would be willing to work with such a compromise, especially if there is no option to turn the NR down or off. 

I believe DJI has used dual gain paths before and combined the two ISOs in their cameras before (e.g. my Mini 5 Pro when in Auto ISO mode) and while my initial reaction was that this looks processed, I don't notice that so much any more, it seemed to help things when I set the camera into Pro mode rather than the default (Auto). I'm talking about still images in the case of the drone though. Is this kind of combination of dual gain signals prone to artifacts (I can see why it might be, but it depends on how well they can get the technology to work). I've happily used two different 32-bit float audio recorder products and not noticed any artifacts in the results despite at least one of the two using dual ADCs and combining the outputs into one signal. Canon use DGO in the C70 and C300III Super 35 mm cinema cameras but not in their recent full-frame C80 and C400 models, is this because they can't make it work or produce economically in 35 mm full frame size, or are there problems with the technology which lead them to use a different design in their newer products?

Although this hand-held gimbal camera might not be my cup of tea (due to the processed appearance of the image, and lack of support for bluetooth timecode and requiring an adapter to take a 3.5mm audio signal), I am interested in seeing if DJI will be bringing out drones with this new 1" sensor, such as in Air 4 or Mini 6 Pro, and if the technology is applicable to still photography.  I find that in my Mini 5 Pro, the image quality on flat days is quite good, but in high-contrast environments such as streets between tall buildings, I find the images quite noisy, and not sure what could be done about it while staying within C0 or C1 classes that can be flown over or near people. I have noticed the drone camera in Auto ISO mode (still photography) seems to select excessively high shutter speeds in order to be able to use both base ISOs (100 and 400, if I recall correctly) for a combined image with larger dynamic range. I might not quite make the same choices myself, as I usually am more concerned with the signal-to-noise ratio in the main subject areas of the image. Anyway, more transparent and manually controllable logic in these devices would be useful. I am not keen on purchasing a Mavic 4 Pro (with the larger sensor camera, presumably with less noise) because it cannot be flown over people even accidentally (lawfully), and in the type of environment where I would be mostly interested using the drone, there are usually uninvolved people. I would be keen to try out a new C0 or C1 class drone with this LOFIC sensor, to see what kind of images it would produce in my applications. 

I think the 20 mm equivalent focal length of the main camera makes sense in the Pocket 4P as it is probably intended to be used frequently to record video of the operator using it at arm's length and to see at least some environment, the wide angle is needed. Also, one can take the 4K video and crop a bit to get a more normal angle of view (40 mm equiv) while still preserving 1080p quality. The sensor area used for this crop would be quite small though, and the image quality can be a concern.  Even in full sensor size (no crop) the images don't show that much real detail based on the videos I saw. I think for social media and news usage this is probably fine for special situations such as videos of action (racing cars or bikes, with the camera mounted on a moving platform, or when the operator walks around the subject to capture it from different sides, while the device automatically keeps the subject in the frame). For quick clips of such subjects I am sure it is a great device, and very tempting. But if the purpose is to make video that looks really beautiful, and where the automatic subject tracking is not a priority, I would still assume larger cameras are preferred.  The Pocket 4P doesn't have an adjustable diaphragm, and the ND filters have to be clipped onto the lenses, and the Auto ISO doesn't work in the high DR log2 mode (if I got this correctly), which means the operator may often be relying on the dynamic range to get acceptable results rather than controlling exposure. A lot of users will probably let the shutter speed vary, leading to not so smooth motion, but this might not be obvious or completely objectionable when viewed on tiny screens at reading distance away. 

 

I am surprised that the integrated gimbal cameras have the options of these tiny devices (230 g) and then the next step up is the Ronin 4D at 4.67 kg. The other options in between include using a generic gimbal and mirrorless camera which can lead to quite heavy setups as well. I would assume there would be space in the market for some intermediate options, such as micro four thirds sensor size with dual lens(cameras) but no interchangeable lenses, that could weight around 500 g or so. Is there something in the market with those parameters? 

[Emanuel]

1 hour ago, Ilkka Nissila said:

I noticed when watching various videos made with the Pocket 4P using my TV that the images had a processed appearance, especially faces seemed overly plasticky in appearance, like they were not of real people. This could be due to image processing (noise reduction & sharpening) and partly due to the characteristics of the lens. However, the dynamic range did seem spectacular and it is only this processed appearance of the main subject that bothered me. 

Have you actually shot with it?

Sharpening, noise reduction and the other image-processing settings can obviously be set to the minimum ;- )

[Ilkka Nissila]

2 hours ago, Emanuel said:

Have you actually shot with it?

Sharpening, noise reduction and the other image-processing settings can obviously be set to the minimum ;- )

No, I have only viewed DJI's and various youtubers' material of the Pocket 4P. I assume everyone will use settings that will make the footage look its best as they would for their normal work, what would be the point of using settings that produce subideal results? Some of the youtubers did mention tweaking the settings to their preferences. A lot of cameras don't have any settings that would reduce noise reduction of their video files to zero, I would not be surprised if this is the case here as well.