The water pass I ran over night in maya is looking good. It was outputed as an alembic cache file of the water going over a section of the wall. I just loaded the water pass into the nuke, then minutes ago did a scanline render sequence into a Adaptive Frame. Frame looks Great, better then I expected. That means I’m good to go with doing the performance capture.
I’ve memorized the scene in my head looking at it a billion times in maya, so atleast the actor/scene problem will be well defined for that scene. I’m also going to memorize the script, I think it will help me on timing for the performance capture. I probably should record the sound in 3D while I’m doing the performance. I’ll have to redo the ADR for it but it could help me on sequence, maybe just one too many thing to think about though.
… Nuke crashed on the render, is 1000 frames, i turned down how many cores run at the same time with a 50 gig of RAM limit for that job. It made it past the frame that crashed, so everything is good (Adaptive Frames are 8Kby4K EXR files with 4 major layers: color,alpha,Z, normals). About 6 hours to do 1000 frames, so I’ll have to think things through a bit more when I do the AF composites (compositing multiple streams of AF frames into one stream of AF).
… crap … went to take a mourning nap, as I was dreaming I was thinking about some of the hard edges I was seeing on some the renders, then how 180 shutter sample (probably 50 samples) would take care of it, then had a wake up moment when I realized I hadn’t turned the GPU motion blur off of the render – which I’m sure is what created the crash. Yep, just checked, i have the distance vector set accidentally. I’ll restart this batch after I made my “dream” changes. [this run will take about 7 hours but should be a lot cleaner]
… hmmm the 8k by 4k frames are taking a long time, I’m thinking of moving to 8k by 2k frames. Also I just thought up a few tricks that would make the 8k by 2k (or even a 4k by 2k AF animated background) be just as good as an 8k by 4k, by moving the camera spherically while on a shallow depth of field lens changing it’s focus through the shot (I can do depth blur and spherical camera moves on the Adaptive Frames, so a background AF could be probaly 4k by 2k and not be noticable if I used these two tricks). I’ll need to use some tricks like this or the scene will never finish, the final AF composite will still be 8k by 4k, but durring the final AF composite to normal 2k frame, the pre-comp AF files don’t all need to be as high as res if i use these tricks.
I btw turned off all heat , and have a fan blowing into the computer room – is 55 outside so with the fan blowing in should keep the room to around 70 – ick … just checked it’s 73 degrees in the room with 86 degrees F out of the back of the racks. I’m moving around a half a petabyte of disk, the backups should be finished soon … so I will shut some stuff off.
hmm hard disk that has the virtual memory for the operating system cache, crashed. I need to convert the alembic files to fbx, abc files are just too big for jumping around (is one multi-gig file for an entire scene shoot). I’m going to kick off that job, then get back to the AF pre comps. … running this … is going real fast … hmmm that ran fast till is crashed. hmmm … i see the job climbing over 175 gig of ram, is for sure bound to crash. I don’t think the way nuke does sampling will work, they must be ram caching the samples then adding them together rather then just merging the samples on each iteration. I’ll have to do it how my dream told me to, duh … why should I have even tried something else. ok … i have the render time down to about 35 minutes, I’m not going to mess with that anymore today and am going to focus on my performance capture system. Workflow is still crashing about every 80 frames, atleast it doesn’t crash on the same frame, so i just do restarts during the day.
taking a walk I remember Deke from the Nuke users group told me to use anisotropic instead of cubic on the scanlinerender node. I just went back and checked and I had it on cubic, I’m switching it to anisotrpic which is how I did it in my high performance testing – hopefully that fixes the crashes (i kind of shot myself in the foot on that one). I have no idea what anisotropic means btw. [that didn’t help on the crashes btw]
I needed to full out open windows in the winter, back plane of the grid was getting to 90 degrees.