Jan Prochazka
04-16-2008, 03:50 PM
I've read that there is a barrier in Mysterium sensor that do not allow go higher than 60fps@3K or 120fps@2K 2:1.
I've also read that it's not the data amount, but rows-per-second barrier.
So I assume that 120 000 rows per second can be read from sensor.
That was essential technical background.
Now conditions: When you shoot something at higher fps you want to slow down movement nicely, right? So resulting picture usually doesn't contain really fast movements, right?
Question: Can we cheat a bit to get better, nicer results?
Answer: We have to try it, of course :sarcasm:
Idea: To read not every row of pixels, but every 2nd or 3rd row. And shift that rows every next frame.
Example: We want result of 4096x2214. We will record 4096x1107 every 2nd line, so in 1st frame even lines, in 2nd frame odd lines, in 3rd even ... and so on. Because we have just 1107 lines to get from sensor we can go up to 100fps! Then pack it using REDcode as 4096x1107. Then postprocess in postproduction to get back 4096x2214. Use line-to-line interpolation and possibly motion detection with previous-to-next frame average. You probably know it - pure magic! :usd:
Sounds familiar? ... Yes, this is old gold interlaced video. :w00t:
But before you start argue at me - did you tried to deinterlace (make frame from every field using interpolation) HD video and scale down to SD? It worked for me excellent. :wink:
Questions for RED Team: Is it possible? Can be sensor tweaked to read every 2nd/3rd line? Is RED One internal CPU capable RED code 4096x1107@100fps?
Is there any other problem? Maybe with bayer-pattern?
I've also read that it's not the data amount, but rows-per-second barrier.
So I assume that 120 000 rows per second can be read from sensor.
That was essential technical background.
Now conditions: When you shoot something at higher fps you want to slow down movement nicely, right? So resulting picture usually doesn't contain really fast movements, right?
Question: Can we cheat a bit to get better, nicer results?
Answer: We have to try it, of course :sarcasm:
Idea: To read not every row of pixels, but every 2nd or 3rd row. And shift that rows every next frame.
Example: We want result of 4096x2214. We will record 4096x1107 every 2nd line, so in 1st frame even lines, in 2nd frame odd lines, in 3rd even ... and so on. Because we have just 1107 lines to get from sensor we can go up to 100fps! Then pack it using REDcode as 4096x1107. Then postprocess in postproduction to get back 4096x2214. Use line-to-line interpolation and possibly motion detection with previous-to-next frame average. You probably know it - pure magic! :usd:
Sounds familiar? ... Yes, this is old gold interlaced video. :w00t:
But before you start argue at me - did you tried to deinterlace (make frame from every field using interpolation) HD video and scale down to SD? It worked for me excellent. :wink:
Questions for RED Team: Is it possible? Can be sensor tweaked to read every 2nd/3rd line? Is RED One internal CPU capable RED code 4096x1107@100fps?
Is there any other problem? Maybe with bayer-pattern?