Lorin Thwaits
October 25th, 2004, 01:22 AM
I am preparing a PC software application that will directly decode the .m2t files the Sony camera creates, and provide a frame server that can be used with pretty much any video editing package. (Premiere, Pinnacle Studio, etc.) To the application it ends up looking just like a standard AVI file. The beauty of this is that there's no lossy interim recompression step, so quality is tops! The drawback is that it's pretty slow to render each frame. I plan on marketing it for pretty cheap, hoping to address the consumer crowd that wants more options with how to use their footage, ranging from full 1920x1080 60fps progressive (using deinterlaced fields to achieve this), down to anamorphic 720x480 or letterboxed 720x360 interlaced, bringing the frame size up to standard NTSC DV. No plans yet for the Region 50 to PAL DV-sized, but perhaps I can make that happen by the time this little project is done.
Anyway, one mode I definitely wanted to include was the progressive frames from the application's interpreter for Cineframe 24 mode. It uses a little field swapping and re-ordering to get things in line. Along those lines, I wanted to post a sample MPEG2 file that I created by having the app feed into TMPGEnc. So for anyone interested in seeing a glimpse of what the 24fps mode can do, here's a link to a page where you can grab a sample 32 meg file made from the dwaterfront24p.m2t file that Kaku had provided:
http://70.56.179.231/cineframe24.html
This footage started life as a 69 meg m2t, and in my app it reconstructs the frames, then resizes if requested using Lanzcos 3. In this case I opted for exactly half the vertical and horizontal resolution, ending up with 960x540. The reason for this specific choice in downsizing is to achieve very nearly a 4:4:4 color space without any annoying artifacts. For every 1.5 pixels in this resulting video, there is one color sample. The result is very clear video. (Still macroblock artifacts to deal with in high-motion and low-light scenes, but still it can be much better than DV.) A strip of black two pixels tall was added to the top and bottom of the image, ending up with 960x544 stream. (This is so that it's evenly divisible by 16 in both dimensions, and thus compatible with more compression options.)
From there I just opened up the frame server's output in TMPGEnc and compressed it, using the main profile and high level, so a 4:2:0 color space.
I wanted to post this out there so people could check out what the CineFrame 24 feature can do. From this you can check out the timing anomilies present in the 60 field per second to 24 fps hack that is CineFrame 24. Each frame is still recorded progressive by the camera, so you don't lose out on any resolution by using the effect. It's just the varying timing between frames that is a little hokie. It varies between 0.0333 seconds between frames 60% of the time, and 0.05 seconds 40% of the time. In each group of 5 frames there's three having 0.0333 seconds after them, and two having 0.05 seconds after them. (Hope all that made sense.) Anyway, for motion it can give a strange effect. At least the two longer duration gaps aren't next to each other... so the gaps go 0.03, 0.03, 0.05, 0.03, 0.05. And it repeats like that forever. It's actually tied into the 15-GOP sequence, so if you jump into it mid-stream, you can know where you are by finding out where the most recent I frame was.
This sample does suffer greatly from being recorded at a higher framerate, making each frame appear very "strobe"-ish. You can still get a feel for what CineFrame 24 motion will be like from it. I'm looking forward to getting some footage filmed at 60fps in CineFrame 24 so I can run it through the app. A little more motion blur would go a long way toward obtaining a better film effect here.
-Lorin
Anyway, one mode I definitely wanted to include was the progressive frames from the application's interpreter for Cineframe 24 mode. It uses a little field swapping and re-ordering to get things in line. Along those lines, I wanted to post a sample MPEG2 file that I created by having the app feed into TMPGEnc. So for anyone interested in seeing a glimpse of what the 24fps mode can do, here's a link to a page where you can grab a sample 32 meg file made from the dwaterfront24p.m2t file that Kaku had provided:
http://70.56.179.231/cineframe24.html
This footage started life as a 69 meg m2t, and in my app it reconstructs the frames, then resizes if requested using Lanzcos 3. In this case I opted for exactly half the vertical and horizontal resolution, ending up with 960x540. The reason for this specific choice in downsizing is to achieve very nearly a 4:4:4 color space without any annoying artifacts. For every 1.5 pixels in this resulting video, there is one color sample. The result is very clear video. (Still macroblock artifacts to deal with in high-motion and low-light scenes, but still it can be much better than DV.) A strip of black two pixels tall was added to the top and bottom of the image, ending up with 960x544 stream. (This is so that it's evenly divisible by 16 in both dimensions, and thus compatible with more compression options.)
From there I just opened up the frame server's output in TMPGEnc and compressed it, using the main profile and high level, so a 4:2:0 color space.
I wanted to post this out there so people could check out what the CineFrame 24 feature can do. From this you can check out the timing anomilies present in the 60 field per second to 24 fps hack that is CineFrame 24. Each frame is still recorded progressive by the camera, so you don't lose out on any resolution by using the effect. It's just the varying timing between frames that is a little hokie. It varies between 0.0333 seconds between frames 60% of the time, and 0.05 seconds 40% of the time. In each group of 5 frames there's three having 0.0333 seconds after them, and two having 0.05 seconds after them. (Hope all that made sense.) Anyway, for motion it can give a strange effect. At least the two longer duration gaps aren't next to each other... so the gaps go 0.03, 0.03, 0.05, 0.03, 0.05. And it repeats like that forever. It's actually tied into the 15-GOP sequence, so if you jump into it mid-stream, you can know where you are by finding out where the most recent I frame was.
This sample does suffer greatly from being recorded at a higher framerate, making each frame appear very "strobe"-ish. You can still get a feel for what CineFrame 24 motion will be like from it. I'm looking forward to getting some footage filmed at 60fps in CineFrame 24 so I can run it through the app. A little more motion blur would go a long way toward obtaining a better film effect here.
-Lorin