I have had great success in setting up DeepLabCut using an overhead camera for 2D analysis of rats in a research lab setting. I’m considering using multiple cameras for 3D reconstruction, and have a question about how to set up my video acquisition. I’m new to computer vision and have a fairly basic question. Because I’m dealing with moving animals, I’m wondering how precisely synchronized the frame times must be across cameras to properly train the network and/or to reconstruct x,y,z coordinates? I’m recording at about 15-20 fps, but there is always some jitter in the exact frame times and additionally isn’t it likely to have some temporal offset between the cameras? Is this the type of thing that can be aligned and accounted for fairly well post-hoc without too much difficulty, or is it necessary (or strongly advised) for me to create a system that will acquire all images simultaneously across all cameras? Are there some general guidelines regarding what is tolerable?
Thank you for any and all advice!