[wellylug] compositing frames

[Pete Black]

>Ok, I have some scripts I can modify here, BTW, do you need to decrease alpha 
>value by the amount of frames 
>1, 0.5,0.5 or 1, 0.5, 0.25 etc ... or if you have 3 frames : 1,0.3,0.3 etc..
>
>  
>
Yes, you will want to change alpha based on the number of layers, and 
potentially the weighting you want to give to each layer - also, the 
blending modes (e.g. add/subtract/screen/multiply) will change the final 
outcome - if you are happy to eyeball a small number of reference frames 
then you can just tweak till it looks right and then apply those values 
across all the frames, otherwise you might want to think about the 
algorithm you actually want to apply - especially since some blending 
modes will give different results depending on the order or compositing.

>I have been reading that results from IM are not terribly good .. I make no 
>assumtpion against image magick, just reporting.
>
>  
>
Now -the biggie - Both IM and vanilla GIMP are bordering on unsuitable 
for your specific application. The basic operation you are performing 
here is the addition of several large matrices - and unless you are 
careful you will run into colour-value banding and clipping issues due 
to the 8-bit-per-channel colour representation they both use internally 
(IM may be able to manipulate 16-bit-per-channel images, but you will 
probably need to compile in support for this).

FilmGIMP would actually be the best free 'interactive' tool I know of 
for this sort of work - since you really do want to be compositing in a 
16-bits-per-channel environment. Most pro or semi-pro compositors 
(Shake, After Effects etc.) will use 16-bit compositing by default.

e.g. if you try adding 2 'light' areas, each of which has an intensity 
of 248-255, the result will be 496-510 - when the colourspace is 
re-scaled to reduce it to 8 bits (colours can only have intensityvalues 
1-255) you will lose definition, and see noticeable banding and clipping 
artifacts. Not good. This may be the reason for your unsatisfactory 
results thus far.

Your technique is similar to those used in creating HDRI environment 
maps, and is certainly a valid way of obtaining more dynamic range in 
the image - however it is crucial that you work with a 
16-bit-per-channel tool or you pretty much might as well just adjust the 
levels in a single 'exposure' due to the above issues.

Hope that helps

-Pete

>>IF you can be more specific about what you want to do, that might help.
>>    
>>
>
>Well When I reshoot each movie reel I end up wit the  3600 * 4Mb of pnm for 
>one reel of 50 ft (the classic little Kodak one in yellow pocket).
>Ideally I'd like to reshoot with various levels of exposure to open up scenes 
>with strong backlight while recompositing, moreover the ratio signal/ noise 
>should improve as well...
>
>Either I trigger once get x Frames_n, composite x in Frame_final_n
>or I trigger once get one frame, reshoot the all reel n times in other 
>folders.
>Advantage of the first method is image stability, adv. of second method : get 
>several levels of exposure, but image may not be so stable (which if happens 
>just crap the all idea down)
>
>Alternative would be to overlay DVs, Cinelerra may do that, but the GUI is 
>just a nightmare...
>Let's say that I am just exploring before I move to windows........
>
>Na .. just kidding..
>E
>
>
>  
>