Open LeeHY-1 opened 2 years ago
Hi, you can follow panoptic or shelf/campus data format and configurations, but we observed cross-camera generalization is currently a challenge, that is evaluating a model on data with different camera parameters from training. You will possibly get very low performance, but this is certainly a good future work.
@twangnh Hi, Do I have to prepare the corresponding carlibration**.json and annotation files like the panoptic sample data? when I test my single video data?
I have built a multi-camera system, and I want to finish the task of 3d pose estimation on my own dataset. Can I use the mvp to do the work and how should I organize the files?