Messages

Some voodoo: http://groups.csail.mit.edu/graphics/CodedAperture/CodedAperture-LevinEtAl-SIGGRAPH07.pdf

Wow.

Say you somehow manage to obtain a Z-depth map of what`s in front of the camera.
In order for a selective focus to be successfully applied, you need infinite depth of field (or near) in the captured photos. Which is what you are getting in the rendered image out of the 3D software.
So the problem with Canon DSLRs, you can achieve great depths of field in very few conditions. Because of the size of the sensor.

First of all a DSLR can obtain very deep DOF if low aperture is used (22).
In my theoretical example however you don't want deep DOF. Instead you want to utilise the shallow depth to obtain the object form at the current distance. Then, by sliding the focus form near to far you would "travel" the whole z-depth. At the same time you would record the sharp area with a edge-detection algorithm. This would build up the "3d-object" that later could be mapped on a picture shot at exact same point, but with full DOF (apt. = 22).

In magiclantern: turn on focus assist and pull focus from near to far:
http://magiclantern.wikia.com/wiki/Focus_Assist?file=Focus_assist_experiment

you will see dots. Use these dots to build up a depth-map for the whole focus-range....
Anyway, all this could be done in post anyway.

I reported an issue about encoder stopping if display goes to sleep.

Your gimbal seem very solid! I bought the came-tv version which works well, but when I load it with 3kg the motors are too weak. What motors did you use and how did you build the frame?

The 70D essentially has phase-detect sensors on every single pixel, making something like this at least theoretically possible (emphasis on theoretically).

Cool. If to be solved I think a recording with a bit of movement is an easier approach. By tracking the footage you can re-create the depth. There are several ways of doing this but this explains one way:
http://research.microsoft.com/en-us/um/cambridge/projects/visionimagevideoediting/unwrap/
However, thats for post-production.

Another way would be to use the focus stacking and take x number of images while focus is going from min to max. Take all the pics, analyse depth with edge-detection algoritm (preferably in the computer). Focus stacking is already implemented.

A third way (theoretically) could be using the already implemented focus-point algoritm and build up a rough 3d-data from the edge detection during a short recording (if combined with a value of focus distance this would be even easier), where focus is pulled trough the z-depth. Maybe something one could do this with the scripting engine?

A circular polarizer is just a linear polarizer followed by a device "scrambling" the waves from linear polarisation to circular.

Thanks for that explanation! Couldn't really find out what that circular thing did. Just ended up in texts about linear polarising filter. Thanks!

I bet it is not! Rather a linear polarizer on top of a circular.

Oh, i didn't know? I have to say that I have never understod how circular polariser would worked anyway. Made me look it up. Seems like all "circular-polarizers" consists of one linear polirized glass and one cir-pol.glass ? Find the whole thing to be quite strange - like quantum mechanics.

I really like my variable ND-filter. Went through 100s of reviews and tests to find a good compromise between price and quality. I ended up buying a Tiffen 82mm Variable ND-filter + some adaptor rings (to use with all my lenses). It was quite expensive, but does not produce any reflections or blurriness to the image.

http://cgi.ebay.com/ws/eBayISAPI.dll?ViewItem&item=360875292783

The filter works as both a polar filter and a variable nd. There is a slight color shift to blue/yellow depending on what angle it is in. Most ND-filters have this property. So if twisting it goes from clear to dark in blue-tint, then dark to clear in yellow tint. The color shift is very subtle and easy to correct. During this twist reflections in windows/water etc changes just as a circular polar filter does. So, this filter is kind of both a ND-filter and a polar filter (actually its two cir-polar filters on top of each other).

Usually I set my lens to a nice apperature (3.2) and then a low iso 100. then i twist the ND until i only have minor areas of over-exposure (ETTR) to gain as much DR as possible.

JxM:
Have you tried to first set the output-directory (W)? I had this problem before in an earlier build where the output directory was pointing to a removed folder.

A good algorithm for debayering could probably also solve the chromatic aliasing because of the line-skipping. Maybe using some heuristic methods for recognising "unlikely" pixel-values. Combining this with some "superresolution" methods (using several frames as a1ex has linked to before) and the result could probably get very good.

Really impressive!! Was glued to the screen for the whole bit. Great work!

Luck to you and your home!
take care

Are you a blonde?
I think its a hair, stuck on the lens blowing in the wind. Do you get it when shooting anything else?

Why 200..? Buy 2 cheap sd - one for mk2 and one for mk3.  put ml on each card and leave them in your cameras.. So you only have to change the much more important and faster cf card.

the mkii doesn't have any sd-card-slot.

Wow! Kept me watching until the end! Great ending, great edit/grading and vfx!

The only thing I reacted on was that it was obvious that the pics were faked. The fact made me wonder if it was intentional or just poor PS skills of the protagonist (but then why so many likes? Wouldn't the online community also see this? . Anyway thats feedback.

Inspiring and impressive

Thanks!
For this test I extracted the frames with the MlRawViewer. Read about some people having issues with strange pattern from the debayering algorithm? Does that apply for MLV from the 5D2 as well? If so, What is the current best method to convert the MLV to DNG/CDNG at the moment?

[May 5th]

Seems like this thread is "dead", but I have just successfully finished a 3 day shoot using two 64GB Sandisk 160MB/s cards and a 5D2, shooting raw at 1856x942. I ended up with about 520GB of MLV-files. I used the May 5th build as I wanted to test it before using it in production.

In total i got around 10 corrupted takes out of 313. 5 of these where early stop frame drop after just 1 sec (?), and so also easy to discover. One or two might have occured while filming in 3x crop mode. Only two corrupted takes had pink frames (just one pink frame), but these two takes was recorded while using an external HDMI-monitor (Peak 5.6" 1280x800) so that's probably the reason. Also one take got corrupted because card got full.

So far the result looks great (I think) Here are some frames graded in LR just for reference:
https://www.flickr.com/photos/poromaa/sets/72157644740100612/

cheers
Joakim

Regarding the sound, try to load the modules named: mlv_rec, mlv_snd  (dont use at the same time as raw_rec).
read more about mlv here:
http://www.magiclantern.fm/forum/index.php?topic=7122.0

Would be great if you had nightlies or compiled versions that one could test of the latest 1.1.5?

It should not trigger automatically while recording.

That is correct. Just tested and it works as intended, so no bug here

Any bugs may be reported here:
https://bitbucket.org/hudson/magic-lantern/issues?status=new&status=open

However, before you proceed make sure you can easily replicated the error. Provide a step by step instruction on how to replicate the error so that the developers can see what you see.

Also provide camera model, build (date/serial number) and module etc.

I have done hard testing on my card on several builds from february to today and had no problems, however I have 5D2, and a different build firmware than you. So its hard to say.

Just a thought - have you tried turning all global draws off and also chose preview: canon (not any preview that uses filters, like the pixely-grayish one).

Also try the different settings in the canon menu regarding video-mode. On my 5D I can choose NTSC or PAL. If I choose PAL and 24fps i get better performance than NTSC 30fps and fps-override: 23.973.
Its like i get some vertical sync problems when using the latter, and once (have not been able to reproduce) I got a magenta bar at the top of the frame.

Also make sure you are not zooming or anything while recording. (there is a setting that turns on 5x zoom if touching the focus ring - turn that off - its of by default though).

If you can find what makes the error you can at least report it.
good luck

have you tried powersave in LV?

There is no dedicated OIS-powersave feature in Powersave menu.
This feature would be most useful between shots, when framing subject or trying different lights. At the moment I will just go with turning off the thing on the lens, but sometimes I simply forget

The mlv_play module is gone in the nightlies for the 5D2? is this intentional?

On my camera 5D2, with Tamron 24-70 VC and the Canon 70-200 ISII, the stabilisation is on the whole time in liveView. Only way to turn it off is to exit liveview (probably a good idea) or turn the switch on the lens (but then I have to turn it on again when filming). Sometimes (often) exit liveview is not good since you might do other settings there.