It would probably be much better for performance if croprotate were to crop the video smaller first, then rotate, and then crop to the final size. This way, ffmpeg will only have to do bilinear interpolation on a much smaller segment of the image, which should hopefully save a lot of computing power.
I'm probably not going to get around to making this change, because the summer is almost over and I have a lot to do, but it could be something to look into doing to make the code run much faster, especially since the croprotate step takes the most time to run in the entire pipeline.
It would probably be much better for performance if croprotate were to crop the video smaller first, then rotate, and then crop to the final size. This way, ffmpeg will only have to do bilinear interpolation on a much smaller segment of the image, which should hopefully save a lot of computing power.
I'm probably not going to get around to making this change, because the summer is almost over and I have a lot to do, but it could be something to look into doing to make the code run much faster, especially since the croprotate step takes the most time to run in the entire pipeline.