About the Med-Tele-Mode of the Mini 5 Pro

Hardware
1

rich text editor image
rich text editor image1891×1332 2.4 MB

I understand that the Mini 5’s Med-Tele-Mode is only a crop on the central quarter of the image, resulting in a 12MP image.

After downloading some images (DNG + JPG) to my computer and opening the DNG in a raw editor [1] I was a bit surprised to find an uncropped 50MP image. The preview thumbnail on the DNG shows the cropped shot, which is not a surprise, but what baffled me is the accompanying JPG, which is a 50MP (8192x6144) JPG! To get the same part of the image (or “tele-mode”) I need to crop the DNG down to 12MP .

Anybody knows what’s going on here? Does DJI use some kind of AI upscaling on the cropped RAW input?

[1] darktable, after conversion with a DNG converter, since libraw currently chokes on the files; different story

(P.S.: I also asked on reddit, but nobody answered. Maybe some of you guys know more :slight_smile: )

2

They both look about the same resolution detail wise, so probably just some clever post processing. The absence of chromatic aberrations is pretty cool

2 Likes
3

It is RawSpeed that doesn’t currently support these. LibRaw should, and one might get by w/o the extra conversion step by temporarily leveraging the unsupported libraw_extensions workaround.

4

I’ve tried with upscayl, which is pretty amazing and gives me similar results. Guess I got confused by the lack of clear documentation from DJI.

Now if only RawSpeed would properly support the RAW files :wink:

5

Hi, thanks for chiming in! The workaround actually worked, I just had to adjust the raw black point. However, I think I’m going to stick with dngpreprocess because that also fixes demosaicing and the OpcodeList3 processing (which is, to my understand, lens correction).

Anyway, I appreciate your effort working on that stuff in your free time. Thank you!

Btw. I tried to upload Mini 5 DNGs to raw.pixls.us, but they never showed up in the repo. Would they be of help after all?

6

Sure, and that’s independent of RawSpeed since some part of OpcodeList3 is not fully supported by dt anyway - the distortions and radial vignetting are, but gain map isn’t IIRC and that’s noticeable for small optics like in this case.

The uploaded RPU samples go through a manual vetting and publishing process (not by me). Since the missing compression feature is well known, having another sample is not critical at this point, more of a nice to have for future testing.

1 Like