Getting RAW Video

[stvnms]

Thanks again! I know all this great effort has been developed with a D5100 but if you need to do some testing on a D3200 I could help.

[stvnms]

What did I do wrong? This is my result of LV_new0.rar file from Photoshop.

Thanks in advice

[nikkor]

This is getting really exciting.

[a1ex]

Please find attached a DNG converter for the above files, based on ML RAW tools.

These images have more than 12 bits - the two most significant ones are missing, and this is why you will get funky artifacts in highlights.

I was able to guess the 2 extra bits in one of the images, as a proof of concept. This functionality is not included in the attached converter though. The guessing algorithm adds 4096 to certain pixels until it gets rid of discontinuities in the image.

Full-resolution image:
LV__0000x.jpg
LV__0000x.DNG

[astronomer]

Great work leegong.

I briefly looked at the histogram from raw bayer channel. There were significant gaps spacing at 17ADU, thus the actual data is only 12-4=8bit.
Maybe there are additional register to scale the data. As a1ex had mentioned the highlight problem, it's possible the digital scaling overflowed the 12bit output.

If you look at the highlight in the other image, all those number are the same revealing the lower 12bit from the image- "3698".

[ron5800]

Great job leegong!

[leegong]

Three major problems need to be solved now :
1) Highlight problem
2) How to activate TFT displaying while dump LV raw . The current patch forces LV raw frame written into RAM ,
so no input for chip4001 now , thus TFT displaying is blank while LV RAW stream is saved to card .
3) Latter dumping in the steam sequence become abnormal .

All attachments in previous thread come from RAM dumping , i didn't modify any byte ,
in fact the dumping address is exactly same as that of the earliest LV raw patch several months ago .
Regarding 1) , my gut feeling and guessing : maybe it's due to special encoding mechanism ,
a NEF RAW container is the best solution EDIT : might be helpful , otherwise we have to use a1ex's DNG container .
maybe there is other kind of problem for shadow , just it's not so sensitive for peoples' perception .
Regarding 2) , lots of hacking are needed .
Regarding 3) , maybe it's due to poor writing speed of my SD card which was very old and cheap .

[leegong]

Here are 11 successive LV raw frames .

[leegong]

I just realized that highlight problem is probably due to clipping in addition mode , see similar description stated in SONY IMX036 manual .
Pixels addition also might be explaination to astronomer's 17 ADU finding .
EDIT : @a1ex , does canon video stream work in sensor addition mode ?

[a1ex]

@astronomer: that's a great hint. Indeed, the lower 4 bits are not random noise, as you would expect from LSBs - they actually seem to be some significant bits, although I wasn't able to figure out their meaning yet.

@leegong: the data is not clipped to white. Rather, instead of say 5000, the DNG file contains a value close to 904 = 5000-4096 (value not exact because the last 4 bits are unknown).

On Canons, 5D3 uses 3x3 pixel binning, and all other cameras use column binning and line skipping. However, that doesn't cause any clipping issues - the data is already scaled down.
http://www.magiclantern.fm/forum/index. ... #msg118232
http://www.magiclantern.fm/forum/index. ... #msg123894