Currently, the Canon VR plugin for Adobe Premiere (and the VR utility) require a jpg or mp4 that is directly exported from the Canon camera with proprietary meta data written that tells the plugin to automatically do Canons equirectangular projection.
This limitation is completely unnecessary and greatly limits the functionality and quality Canon VR system. For example, if you hack the metadata for a non-native file correctly, you can trick the importer into processing a non-canon native shot, and it does so just fine. This is currently the system that I use to pre-process my shots but it is very cumbersome, requiring me to render, transcode and re-import my VR shots, wasting time, quality and hard drive space. It also makes shooting RAW a non-starter for video and it makes it so that for RAW time-lapse stills I have to lose bit depth by going down jpeg and then wasting time hacking the metadata for my jpegs.
The reasons for wanting to pre-process video are numerous. Clog3 8k footage is very noisy and often suffers from chromatic aberration that differs in location between each eye (lens). This has disastrous effects on stereoscopic vision making many VR shots unusable. Ideally, the CA would be removed before equirectangular projection. Ideally the noise would also be reduced before stretching out the pixels.
The ideal solution would be for Canon to just make the importer more simple, and instead of relying on metadata, just create a "Canon VR projection" effect and let the user tick some boxes to tell the effect what kind of VR footage it is looking at. It would be nice if the effect also had CA removal built in like the Canon RAW importer. Because each eye is a separate lens centered on one half of the sensor, it makes manual CA removal more difficult than with a single eye shot. This would fix all of the above problems.
*On a side note, I wish the Clog3 8k footage was just plain less noisy as it is a must for shooting in natural lighting...it is orders of magnitude worse than 4k HQ.
