I’ve got 120,000 stills, Canon 5D Mark II raw CR2 files from a 5 camera time-lapse rig to process for my IMAX film by the amazing Colin Legg who donated this all to the film. It’s 12-day, 12-night time-lapse. You can see the rig in action here: https://vimeo.com/31068829

I’m trying to ponder a workflow for this that makes sense and will not cause problems with shifts over the process. About 10% of the footage has issues from missing frame from 1 to 5 cameras, bad exposure due to fault. Here’s my rough of the steps. For noise reduction I’ve determined that RevisionFX is the only one that does not band the very subtle low contrast shading. Both Red Giant and Neat Video do band regardless of settings or bit depth.

The goal is to give the stitching program clean, usable images.

1.       Build raw sequences in AE in 16-bit.

2.       Apply Noise reduction – however, since AE is not aware that the images will be stitched, noise reduction may create unevenness at blend points unless REvisionFX DE:Noise can somehow be controlled. The other option is I also have Noise Ninja will has built in camera profiles specifically for this camera and could batch process prior to AE but that is a really, really big batch job.

3.       Apply fixes (fill gabs, fix artifacts – everything from spiders to exposure)

4.       Deflicker? – not sure where it’s best to apply this step here as flicker may vary. I’ve not use Deflicker much on long timelapses.

5.       Export as 16-bit TIFFs.

6.       Stitch in PT GUI Pro, export as 10k by 10k TIFF.

7.       Bring back to AE for correct and comp work

You can see Colin’s rough test attempt that reveals some of the issue – it’s contrast enhanced for artifact reveal:

> I've done another render, this time from the last day covering sunset -> sunrise. This covers the period when the cameras were bulb ramping. Outside of this I had to use manual exposures (no choice) and they suffer the classic 1/3 step flicker and beyond 1/60 s, shutter flicker.
>  
> http://vimeo.com/35544430  pw - lakedorasunsetsunrise
>  
> Still using the same denoising first workflow so banding is still bad. Also note the red splotch at top....this is reflection in spider webs from red status lights on the other cameras.

Due to artifact and variations over 120,000 stills it’s hard to spot check and be sure all issues are known. I need some reliable workflow that won’t be thrown off my problem with single or batches of images.

I do have time markers for what sequence don’t have gaps or misfires.

Thoughts?

Admittedly I’ve not done a lot of motion tracking.

 

I’ve got a shot—head and shoulders interview on green screen (720p 23.98)—where I failed to notice the little white speck at the edge of a nostril.

 

Steps done so far:

1-      Exported the first frame of this clip to Photoshop>rubber-stamped/fixed the white speck area, then made a small selection of the immediate area (with 2 pixel blur)>created and saved to new layer>deleted background layer and saved as .PNG file.  So my “repair” area is a small patch, also 1280x720, surrounded by lots of transparent area.

 

2-      Imported the bad clip.mxf (P2 video file) and the patch.png file to Ae. Turned on tracking so I’m tracking Position and Rotation. Set the two tracking points to 1) exactly on the white spec and 2) a point on an ear> defined the target to be the patch.png file and ran the Analyze function to set the related keyframes.

 

During his interview, as you might assume, there’s quite of bit of head shaking. The nose goes all over the place. There’s motion blurring that occurs on the clip because of the motion, and that seems to shake the white spec tracker loose. It drifts around slightly.

 

But here’s the deal. After applying all the motion and rotation keys to the patch.png file, the patch locates itself (according to the keyframes) several inches away on his cheek. My nostril repair patch now appears to be a mole that pretty closely tracks to his cheek rather than tracking to the nostril area where it was located at frame 1 when the tracking began.

 

My question is:  what is the trick to take a set of motion/rotation keys attached to the patch.png file, and slide…

 

…uhhh…I got it! Change the Anchor Point! Doh!

 

OK. So now that’s solved, have I hit Ae’s wall, in that its motion tracker is only so good in high motion areas?  Any other options to pin that tracker down tighter?

 

Thanks,

 

John

 

 

 

This message, and any attachments, is intended only for the use of the addressee and may contain information that is privileged and confidential or otherwise exempt from disclosure under applicable law. If the reader of this e-mail is not the intended recipient, you are hereby notified that any unauthorized use, dissemination, distribution or reproduction of this message is prohibited. If you have received this communication in error, please notify the sender by return e-mail and delete this e-mail and any attachments from your system.  ­­