supplierdeeply

500d development

  • 2487 Replies
  • 320718 Views
Re: 500d development
« Reply #160 on: 13 / February / 2011, 23:10:49 »
Advertisements
how would I go about 'getting this on my camera, and will this enable manual functions for ISO Aperture and Shutter speed while recording video?
Frazer.

*

alins

Re: 500d development
« Reply #161 on: 14 / February / 2011, 01:05:46 »
how would I go about 'getting this on my camera, and will this enable manual functions for ISO Aperture and Shutter speed while recording video?

this page should explain how to get ML on your camera: http://magiclantern.wikia.com/wiki/500D
to answer your second question: not yet

*

alins

Re: 500d development
« Reply #162 on: 14 / February / 2011, 01:30:14 »
I'm releasing a new build with the most notable change: zebras. I've been playing a little bit with zebras and i like what i'm seeing. Huge kudos to RoaldFre for making this happen.
 
I want people to start playing with zebras, see what works what doesn't work. I don't want to give the wrong impression here that zebras have been nailed down, just that they start working and i would like to spread the joy :)
Other change in this build that jumps in mind is a little fix in HDR where in manual mode all the shots where pretty much identical (same shutter speed same aperture). Now the shots are taken correctly by actually changing the shutter speed.

Changing the ISO doesn't seem to send the camera in a loop anymore. One issue that i believe should be mentioned: while recording a movie i tried changing the ISO and the camera crashed with error 70. Same problem if i try to change the shutter speed. If you see this error, just take the battery out and restart.

If you find any problems please check the issues page first: https://bitbucket.org/alins/magic-lantern-for-500d/issues and if there's no issue already open feel free to create one. I will like to start using the issues page, i believe it communicates better what are the current problems and it helps the developers to prioritize what to fix.

*

alins

Re: 500d development
« Reply #163 on: 14 / February / 2011, 01:52:09 »
Alex, my latest changes are in bitbucket if you want to diff them


Re: 500d development
« Reply #164 on: 14 / February / 2011, 02:17:30 »
Thanks Alins, you can be sure I'll be testing it out :)

Re: 500d development
« Reply #165 on: 14 / February / 2011, 07:05:07 »
I just tried your latest build, Alins. The zebra's seem to be working perfectly here! (at first sight)

I've added the 500D VRAM info on the wiki: http://magiclantern.wikia.com/wiki/VRAM/500D
I will try to find out if those sections change when in (different) movie modes next (obviously the 0x4FCB6C00 section remains there, as it's what we use for zebra :-)).

*

Offline a1ex

  • *****
  • 671
  • ML dev
Re: 500d development
« Reply #166 on: 14 / February / 2011, 07:17:18 »
This one looks pretty interesting: http://i.imgur.com/arQyU.png

Could it be a method for contrast detection? (for autofocus)

I mean locally swapping the pixels would generate high contrast in focused areas and little change in out-of-focus areas.

Re: 500d development
« Reply #167 on: 14 / February / 2011, 07:30:01 »
I think that's a bit too big of a hack. For instance, just reversing the endianness and looking at how big a difference you get will fail to detect contrast in horizontal edges (as you are only swapping left-to-right, so only vertical directions get scrambled).

If you also swap the pixels vertically (say, swap them in every 4x4 block both vertically and horizontally), then you'd loose the sensitivity for diagonal edges. It could work if you actually "rotate" the block fully (that would only fail on some pathological cases where your scene is invariant under such rotation of 4x4 pixel blocks)

Overall, I'm pretty sure there are better alternatives available.


*

Offline a1ex

  • *****
  • 671
  • ML dev
Re: 500d development
« Reply #168 on: 14 / February / 2011, 07:53:41 »
Of course.

In my offline tests I've got very good results with morphologically bluring the image (open and close) and checking where fine detail is lost. This method is much more robust when it comes to high-contrast OOF edges, but it needs to analyze a 5x5 region around a pixel in order to detect the local "focus strength".

Plain edge detection (which simply checks difference between adjacent pixels) works well in many cases, but has a lot false positives in certain situations (OOF high-contrast especially).

These methods are used in focus peaking in the latest builds. The morphological method uses a very crude 1x4 "area" instead of 5x5, which makes it very sensitive to noise.

I've got similar performance (i.e. very good) with median filter (again, with at least a 5x5 area). If you know how to implement it fast, that would make focus peaking much more robust than it is now on 550D.

*

alins

Re: 500d development
« Reply #169 on: 14 / February / 2011, 10:19:57 »
yap, zebras are working really nice :).
again, awesome job RoaldFre.

 

Related Topics