Idea for speed increase

Started by primemultimedia, June 20, 2013, 09:02:58 AM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

primemultimedia

Hey all not sure if this has been brought up before, I didn't even now what to search for to see if it was. Basically I had an idea to try to squeeze a little more resolution / speed out of the slower cameras. Is it possible to alternate frames between low and high resolution? So for example frame 1 would be 1920 and frame 2 would be 720 and so on, kinda like how GOP works I suppose except using resolution changes. Then in post we would upscale every other frame to match.

Just throwing it out there. If this is not at all possible please remain calm and continue coding  :D

a1ex

You could try one with line skipping and one without. Or, interlaced.

Before spending development time on this, you have to figure out the postprocessing, see how bad is the aliasing and motion artifacts introduced by this and so on. Without some serious math on the postprocessing side, I think it's a waste of time.

primemultimedia

Thanks Alex, I might do some experimenting then on the processing and see how it looks  :D

larrycafe

raw2dng could combine the interlacing frame into full resolution frame.

I don't know if video editor can import interlacing frames otherwise, I am currently using Edius

a1ex

Please see reply #1.

It's your job to do that.

larrycafe

sorry for throwing an idea which I don't know how to make it for real.

any tools which can help me to make a full resolution frame becoming interlace frame?

and I will need to combine the interface frames back to some full resolution frame. if you want some text illustration or showing the concept with some simple 5X5 matrix, I can do make it.

eatstoomuchjam

Quote from: larrycafe on June 20, 2013, 07:17:14 PM
sorry for throwing an idea which I don't know how to make it for real.

any tools which can help me to make a full resolution frame becoming interlace frame?

and I will need to combine the interface frames back to some full resolution frame. if you want some text illustration or showing the concept with some simple 5X5 matrix, I can do make it.

Combining frames sounds like a complete mess if anything in the frame is moving unless you some pretty sophisticated motion estimation software to go with it (similar to what compressed codecs do with motion estimation).

You should be able fairly easy to demonstrate this concept in action without motion estimation.  Take an existing dng sequence and use lightroom to export every even-numbered frame as 1080 and every odd-numbered frame as 720.  Then resize the 720 frames back up to 1080.  Then import into your video editor of choice.  It's not complicated or challenging/difficult.  It's just very time consuming and (IMO) not likely to be worth the effort.
I'm not even sure which cameras this would help - the Mark II can already keep up with 972p at 16x9 (looks downright decent when blown up to 1080) and even the 50D can do 896p.  I have to assume that both of those look better expanded to 1080p than a weird mishmash of 720p and 1080p...  and most of the SD cameras can only handle like 540p so at best, having half of the frames at 720p is just going to buy you another fractional second of recording at 1080p.

Oh, and additionally, unless I'm mistaken, it's not easily possible to alternate 1080p and 720p in camera while recording unless you want the 720 frame to be nothing but the center of the 1080 frame.

larrycafe

for 60D, it is now doing 1728 X 1152, maximum 108 frames.

if the writing can be interlaced, each frame will be 1728 X 576, it should be able to capture more frames.

and yes, the problem is make a proof of concept, and it is only benefiting a very small amount of user who use 60D, with larger buffer but slow write speed.

otherman

With avisynth scripts it's possible use interlaced frames to get perfect motion estimation and retain all the details. With a full resolution frame, and two "half" frames, you can get 3 frames at full resolution, without visible loss, at half of datarate. It can be very usefull for SD only camera, like the 550d. I'll run some tests with avisynth, but I'm pretty sure it can work.