I was given a 360 video of 12288 x 6144 dimensions. I separated it by frames, gave it some treatment to turn up the shadows, then converted the spherical images into cubemaps (3911 x 3911).
It sucessfully aligned the images, however the produced model is quite horrible.
What should I do to get better results?
Hi, the models from video are still worse than from images. And these images were created from 360 video, which is basically not supported by RealityCapture, so there could be some errors.
How many features do you see on separated images? Are you using georeferencing? What are your reconstruction settings?
According to your screen shot the image is quite blurry, so it also could have the influence on the model quality.