This repo provides streamlined scripts to obtain 6D camera poses from videos, even the ones with dynamic objects. The method comes from EPIC-FIELDS。
Follow instructions in the EPIC-FIELDS repo.
get_pose_any_videos.py is the main script used to get poses from any videos. It entails 3 stages:
- Select sparse frames from homography
- Reconstruct sparse frames
- Register dense frames
It assumes a required argument input_video_frame_paths which directory contains a set of images (jpg or png). The output will be at the colmap_models folder with COLMAP output format. Additional information can be found at logs, sampled_frames.
Use check_pose_with_gt_dycheck.py. Do remember to change some paths including DATA_ROOT, SEQUENCE and line 143.