Hello, I have been digging around the internet for a solution for recording field sports with a single stationary unit which led me to here! I downloaded the application and the test videos and stitched them together by following the guide and all seems to be working.
The issue I am having is that I am trying to use the 2 gopro hero 9s I have. I have downloaded the npz file for the 9 and I have added a key to the registry. However I cannot seem to get the stitching to work properly. My test footage was taken inside a large room with the cameras about 60degree angle from each other. So the video has an overlap of about 1/4 of the screen, The settings are identical at 4k 60fps. No hypersmooth.
There are duplicate objects in the stitched video and it just does not look quite right.
My questions are:
What are the expected settings for the 9? Is 4k @ 60fps okay or should it be 5k @ 30?
Does the distance of the objects in the video matter? Is me testing inside a large room a problem?
Would I get better results if I adjusted the angle of the cameras?
Thanks for any assistance you can give!
This product seems awesome and if I can get it working I will happily fork over the money for the license.
Your videos should be recorded in 5K wide mode with Hypersmooth turned off. If you still run into stitching issues with these settings, please send me the videos via swisstransfer.com and I’ll help diagnose the problem.
Thanks for the response! I have retaken the videos outside with the recommended settings and the program seemed to be fairly successful at stitching these together…I will need to try it with some more realistic footage but it looks very promising!
With that, I kind of went down the rabbit hole of trying to learn how to do this type of stuff myself with my own python scripts. Would you have any interest in answering a few questions about the algorithms and logic you have on going on behind the scenes?
Mainly on the actual feature detection and stitching/blending. I have calibrated my cameras own cameras with the fisheye model…although that does not seem to be the model you used with this hero9 npz file. But I have my videos undistorted from the wide fov I recorded in. I am just having trouble stitching them together…If you would be interested in sharing some knowledge about this I would love to ask some questions or potentially chat.
And another issue I found with my stitched video is that the resolution seems quite poor. The stitching worked fairly nicely but when I play the video back it seems to have lost a noticeable amount of detail. Any ideas?