Editing AVCHD, problems with proxy

Testing footage from a Canon HF200. I was surprised that the Sequencer recognized the .mts when I added it to the timeline. I could get choppy playback, with audio, at the full resolution but when I set the proxy to 50 (and then 25) playback did not improve. the proxies did seem to render properly, with some slight shearing on some frames but the playback stutters.

Are there any tips from more experienced Blender editors?

Ubuntu Jaunty 9.04_64 on Quad-core cpu with 8GB ram. Blender 2.49 svn.

I can tell you this: You’ll save a lot of time and hassle if you convert the video file to a sequence of images (jpegs work fine for AVCHD stuff). It will read the individual images much faster than it can track to the correct spot on the compressed video. Just delete the images when you’re done with that shot, and keep the source video file, to save space.

I hadn’t thought of that and I’m not too keen on the idea but shouldn’t the proxy feature do this for me?

My goal is an easy way to edit this particular content. I wanted to avoid intermediate movie files and hoped the proxy button was the answer. I’m currently researching options for transcoding using ffmpeg or mencoder. One of the issues is telecine conversion, or pull-up. My first test shot was using the 24p setting, tomorrow I’ll record a 60i. These Canon Vixia’s don’t really shoot 24p or 30p… well, i guess the sensor is scanning at that rate but it’s still recorded in 60i. Yuck.

Found HandBrake for linux but the conversion is taking forever. (~1 fps on 1.8GHz pentium-M as opposed to minutes with mencoder.)

Convert to image sequences is the way to go.

Try AviSynth + VirtualDub?

If you don’t convert to images or intermediate lossless codec then any edits etc you do stand a good chance of being compressed and recompressed each change.

One of the issues is telecine conversion, or pull-up. My first test shot was using the 24p setting, tomorrow I’ll record a 60i. These Canon Vixia’s don’t really shoot 24p or 30p… well, i guess the sensor is scanning at that rate but it’s still recorded in 60i. Yuck.
It’s progressive in a 60i container and I don’t know if you’ve been over to www.hv20.com but you’ll find some very nice Avisynth scripts to ease the progressive out of the container without transcoding/reencoding, you really don’t want to transcode, there’s scripts to do the pull down for you as well.

Recording interlaced is of benefit when you want to do slow motion, it looks far better than doing slow motion with progressive footage.

Just rereading your intial post, by using proxies you are using jpg image sequences I think, they are stored in a sub folder to where the movie files live, so playback of the jpg image sequences should work fine.

So is this a sound issue not image? Have you tried a 2.5 build from graphicall using the new sound system, just as a test?

I’m on linux so I can’t use Avisynth, otherwise yeah, it’d be the way to go. Thanks for the info and link. Got some reading to do…

Here’s what I’ve done to overcome the artifacts from the mts in Blender:

First I build ffmpeg (from svn) using these configure options…

./configure --enable-gpl --enable-libmp3lame --enable-libxvid --enable-libfaac --enable-nonfree --enable-libx264

Then I transcoded the mts:

ffmpeg -i 00000.MTS -vcodec libxvid -b 18000k -acodec libmp3lame -ac 2 -ab 320k -deinterlace -s 1440x1080 00000.avi

Much better playback performance but I need to compare (difference) a few frames from each to see how bad the loss is. Again, though, proxies still do not seem to perform correctly. I’ll have to investigate this more in 2.49svn.

I only quickly checked into the sequencer from a summer of code project and didn’t see anything usable. I guess I need the main 2.5 branch or is there a different branch focussed on Sequencer development?

Some rather depressing reading on the hv20 forums. Anything linux related and you’ll get comments from this Debbie-Downer of sorts. Maybe it is time to invest in Sony Vegas and XP… or is it Win7?

I’m on Linux too, Ubuntu and I use AVISynth + VDubMod + TMPGEnc 2.5 and 4.0 Xpress + DGIndex all under wine.

I use blenders Frameserver, with a blender vfapi (compiled with Ming) and ran using a .blu file to frameserve to TMPGEnc as set out by Peter Schaile in the wiki. No problem. Perhaps try it. :slight_smile:

I hate the idea of transcoding anything due to the risk of quality loss, MPEG4 ( AVCHD) to MPEG2 (.mts) has got to be lossy?.

With VDub + AVISynth you can export out to lossless AVI. You mentioned about not liking image sequences but if it was all a 3D project rendered out then there would be large amounts of images anyway. I put each vid clip image sequence in it’s own folder within the project folder to reduce directory listing delays from masses of image files.

I’m avoiding using ffmpeg with blender as well as I can’t tell what it’s doing or what blender is doing.

Like does Blender+ffmpeg do a YCrCb to RGB conversion for the sequence editor, if it does does it use PC.709 or Rec709 or worse Rec601, does it scale the 16-235 to 0-255 and throw away the Super Whites above 100% IRE (ie 235) in the YUV to RGB conversion. I’d like to hold onto them until I decide what I want to do with them.

From the tests I’ve done it looks like a Rec709 conversion importing a video via ffmpeg and when I compare the same frame imported as a png from AVISynth using PC.709 in blenders vectorscope the ffmpeg import appears to have less information than the png frame ie there are a lot more holes in the distribution of pixels.

But perhaps it’s that ‘a little bit of knowledge is dangerous’ situation and I’m making something out of nothing again. :slight_smile:

So until I’m sure I use AVISynth to do the conversion to RGB where I can decide what matrix and what to do with the levels, I then export from VDubMod as an image sequence so ffmpeg import into blender is not involved, do the editing, up to 32bit float if I’m comping and CC, then frameserve out to TMPGEnc avoiding ffmpeg on the way out as well.

I’m not saying ffmpeg is no good or doesn’t work right, it’s just it’s all hidden from view, there’s no import options.

It’s a bit the same with blenders deinterlace button, what type of deinterlacing? Is it the best, where’s the options?. Would I find AVISynth better at deinterlacing, that’s what I’m using when I do 50i, I mostly do 25P and 50i when I want to do slow mo and such effects.

Although I like choices and to see what is going on under the hood. I could be totally wasting my time converting outside of blender. :slight_smile:

I only quickly checked into the sequencer from a summer of code project and didn’t see anything usable. I guess I need the main 2.5 branch or is there a different branch focussed on Sequencer development?
Well there’s the 2.5 colour management thing so that might be a benefit if it extends as far as the sequencer. Also found a patch for graticules? for the Luma Scope and RGB Parade. I’m waiting to see what comes with 2.5 there are many improvements needed I think, including the Histogram it’s useless as it is and doesn’t work in 32bit float. The scopes are low quality better with the patch. Better CC options as well I think, Xat may have something for that.

Some rather depressing reading on the hv20 forums. Anything linux related and you’ll get comments from this Debbie-Downer of sorts. Maybe it is time to invest in Sony Vegas and XP… or is it Win7?
Sorry about the link, yes it’s a fantastic site, but they have a point, blender’s VSE isn’t a NLE but I think there are a lot of NLE stuff that is pretty redundant or low priority to me anyways, plugins and filters and issues with having to use intermediate codecs, out of sync auido etc, it’s not all plain sailing for NLE users either.

Good Luck.