The input video thumbnails are slow to update when moving on the timeline - frequently done when wanting to confirm or refine sync based of a flash frame.
ffmpeg can generate 1/10 size mjpeg copies of 2.7K MP4s at 90 fps on my machine. These are very responsive and initially appear to work fine (but autopano hits a snag and reports no video present when I attempt to sync on audio, I'd been hoping to then simply switch-out the paths in the .kava file).
1) Please consider pre-generating the thumbnail videos with (intraframe compression) to make the Input Videos tab more responsive. Here's the command I use: ffmpeg -i INPUT -pix_fmt yuvj420p -vcodec mjpeg -f mov -vf scale=iw*0.1:ih*0.1 OUTPUT
2) Please also provide sync adjustment controls next to each thumbnail (similar to the existing widgets with absolute value and +/- controls would be fine)
3) When automatic sync fails on only a few cameras, flag those as red and the others green after applying the sync values - presently it leaves us in the dark. If there are possible complications of the remaining clips having less head, just ripple all the other values as we'd adjust them.
4) Cue to the first frame when auto syncing succeeds.
5) Present multi-track audio waveforms as others have requested
6) Auditory confirmation of sync would be *EXTREMELY* useful in saving lots of time, please provide a feature to simultaneously play all audio tracks for maybe 5 seconds around the current playhead position. It couldn't be to hard to use an off-the-shelf tool to mix it down if necessary. If we hear an echo, perhaps we could then have checkboxes to help us isolate the offending clip. I can't stress how much time this could save.
7) OT given the subject, but on the stitching front it would be great if we had the ability to anchor translation and rotation of a given camera to a nominated point - one we've already aligned - we'd then be able to man-handle the camera to make a best-fit on one other point we consider a priority given the compromises that may have to be made.
Having doubts about the validity of "Accurate Sync" being reported, I'm now resorting to a python script that generates my lowres intraframe mjpeg clips and launches each sequentially in Quicktime while asking for the flash sync frame number before moving onto the next, it then prints the values less the minimum for me to manually type into autopano. All of this manual sync functionality could/should be built-in.
Having said all that, my primary request for Autopano Video is that it become frame accurate and not randomly slip frames on random cameras when rendering (and giving different results even if you just change the output codec). Until then, it simply couldn't be part of a serious pipeline. We're all striving to shoot at high frame rates for smooth motion, and these stutters are unnecessarily being injected into our stitches.
I must say I'm very impressed with the stitching speed you've achieved and look forward to the evolution of this tool.
You most definitely know your job - Kolor devs, these are great points. +1 on all of them, totally. Synch is at the moment one of the most important things in stitching and AVP still leaves a lot to be desired.
Point #7 is not comprehensible the way you've written it, tho. Can you rephrase?
All pro software that lets you pan, scale and rotate images allows you to move the center/anchor/pivot of that movement. For example, rotation typically defaults to spin around the center of the image. Here is a video of the basic concept: https://youtu.be/r9iw7tDfzKw
I've got a stereoscopic shot where I'm happy with the alignment of something in the extreme foreground at the top left of a pair of cameras, but the background alignment could be improved if I rotated one of the images. I want the center of that rotation to be on the relatively well aligned FG object near the top left of frame so it doesn't move.
Basically, this feature would let you pin any point of a camera down to a specific point while letting you move other parts of the image with scale and rotation.
At present, fixing something in Autopano by making alignment alterations has me very worried about breaking something else.
I'll add an additional point too:
8) I'm finding that rendering frame sequences is inordinately slow compared to rendering mp4 or cineform, 0.1 fps compared to 4-7 fps. Why is there such a performance hit when rendering frames?