Animator On Linux

As my equipment gets better and better I am finding the defaults of the crunchbang++ to be a little limiting. One of the ones that was bugging me is the default size of the terminator window that pops up when I want access to the Terminal.

The fix? edit the ~/.config/terminator/config file

Under [layouts] you just need to add the line:

size = 960, 720

I was lucky enough to catch DingTo on IRC #BlenderChat and he was able to walk me through how to get the new “Adaptive Compile” to work in the latest version of Blender. It took a little bit of doing but once I got it working I was more than amazed at the memory optimization of this new feature.

If you would like to try it yourself, and you happen to have an nvidia GPU on a linux system, just download the very latest version of Blender from https://builder.blender.org/download/. (It doesn’t work yet for the Windows or OSX builds.)

You will also need to install the nvidia-cuda-toolkit. This is so that Cycles can compile before you render each new scene. To do that on my Debian Testing machine I just. . . .

sudo apt-get install nvidia-cuda-toolkit

I’m sure you can find instructions for your specific Linux distro if you are not using Debian.

I had to revert back to the Debian Nvidia drivers instead of the ones from Nvidia directly to install this but that’s a pretty painless process with smxi.

So once you have those things working, open up version of blender you downloaded and with the mouse in the 3D view window, hit the space bar and search for Debug. There should only be one option. Select that. Then enter 256 as the debug value and press ok.

What that does is enable a preference box in your interface that will allow you to control different optimizations made for Cycles. “Debug”  Make SURE you are using the Cycles render engine or you will not see this option.

Enable the Adaptive Compile flag.

One extra step that you need to do right now that we won’t have to do in the future is to delete the pre-compiled Kernels for your system so that it knows to compile you a new one.  To do that you go into 2.77/scripts/addons/cycles/lib and delete all of the .cubin files.

Now load up a scene and hit F12 to start rendering. The first time you render the scene it will take a while to compile the new custom version of Cycles for your specific scene but once it is through, it will start rendering your scene. You can go ahead and stop stop the render and start it again. This time it will render your scene without having to compile Cycles again and will use your customized version of Cycles built specifically for your scene!  How awesome is that!?!

It was SUPPOSED to render faster than the standard Cycles but I didn’t find that to be the case with my system. What it DID do was dramatically decrease the memory footprint of my scene. I went from using 3368MiB of V-Ram to only 1471MiB!   The best way to tell how much memory your scene is using is to open up another terminal window and type in nvidia-smi while your scene is rendering.

You will then get a report that looks like this.

So here you can see that I have two video cards in my system.  One a GTX 970 and the other a GTX 980 Ti.  The 980 is the one I was rendering with when I ran nvidia-smi so you can see that it is pegged at 99% usage.  But the thing you want to look for is the “Memory-Usage”  For my scene it was 1471MiB / 6143MiB. So that’s it!   On the BMW27.blend benchmark scene, with blender 2.77a, my memory footprint was 2668MiB.

With the new Adaptive Compile feature, my memory requirement for the exact same scene went down to just 600MiB.  That is a MASSIVE drop.

Every now and then I need to convert a sequence of images into a movie so that it can be given to clients in a more manageable format.  The way I have been doing that lately is with ffmpeg.   It should come standard with your linux install but if it isn’t I’m sure: sudo apt-get install ffmpeg Would do the trick. So the syntax that is working for me is: ffmpeg -framerate 15 -i file_name%4d.png -vcodec png file_name.avi

Breaking it down a little bit: ffmpeg starts ffmpeg

-framerate 15 sets the frame rate to 15 fps.

-i filename%4d.png designates the image sequence, the %4d is the number of units in the numbering of the image sequence.  %4d = 4 numbers.  So file_name0001.png, filename 0002.png etc.

 -vcodec png = setting the video codec.  I have mine set to png here but there are a lot of different ones you can use.  For my needs png works. 

python3 -c 'import os; [os.rename(file, str(file).replace("old_name", "new_name")) for file in os.listdir()]'

In my switch to linux there have been a few applications that have been missing from my tool bag.  One of them is a robust screen recorder that allows me to record time-lapse video.  It is so much easier to work with a client when they have an inkling of what I have done for them. Recording a long session and sending it to them in a super sped up movie saves a lot of explaining. (They usually still have no idea what I have done for them but at least they can see that stuff is happening).

 In my search I tried a variety of screencasting tools.  VLC, screenkey, recordmydesktop, vokoscreen etc.  Of all of those I was most impressed with vokoscreen.  The interface was very intuitive and it the webcam/audio integration "just worked".  Unfortunately it did not have any way of allowing me to set the recorded frame rate lower than 1 frame a second so I had to abandon using it for this need.  I will for sure use it for any tutorials I make in the future though.

After a little poking around I noticed vokoscreen was using avconv to do all of it's encoding so I read up on how I might do what vokoscreen does only with the command line.  After a little while I got something that worked perfectly. So here it is.   avconv -video_size 1920x1080 -framerate 1/10 -f x11grab -i :0.0+0,0 -pix_fmt rgb24 /path/filename_%05d.png

avconv installs by default with most linux distros.  If you don't have it pre-installed I'm sure you can use your package manager to grab the latest.  On debian desktops you can type:

sudo apt-get install libav-tools 

Once you have it installed you just open up a terminal window and type in avconv with all of the variables.

I'll break them down:

-video_size 1920x1080  That is the screen resolution you would like to record at.  This is usually your screen resolution if you are recording your whole screen.

-framerate 1/10

This is the frames per second that you record at.  If you set it to 24 it will record at 24 frames per second.  Here I have it set to 1/10 so it will record 1 frame every 10 seconds.  Perfect for time-lapse. -fx11grab

This tells avconv that the source of what you are wanting to record is your x11 desktop. -i:0.0+0,0

This tells avconv that the area that you would like to record on your screen starts on the top left and that there should be no offset.  If you wanted to record a smaller section of your screen you would use pixel values to offset the X and Y values here.

https://libav.org/avconv.html#x11grab

-pix_fmt rgb24

This tells avconv what color space to record in.  By default it uses rgb32 when exporting a png sequence which includes an alpha channel which we don't need.  With this option set there is no alpha channel in your output. /path/filename_%05d.png