Blender Mesh

Blender Normal

3d Gaussian NVS

Blender Points

4d Gaussian NVS

1. Import mesh or GS points in blender, and set camera track (see setCamTrackInBlender.md)
2. Run visEngine/blender/export_cam_track.py, to export cameras for all frames. These are stored in camera_track_opencv.json
3. Read camera_track_opencv.json in any project, please refer to demo_track_read.py

install the following packages

1. basic

• plyfile: for managing point cloud
• numpy
• skimage

2. ace visualization

• torch
• trimesh: for meshing camera traj
• pyrender: for rendering
• Pillow: for PIL Image
• scipy
• matplotlib
• skimage
• bisect

3. mono depth

• torch
• kornia: for mono depth to pcd
• Pillow: for PIL Image
• opencv-python: for cv2
• open3d: for point clouds

cameras are stored in camera_info_opencv.json with following formats:

• P_c2w: camera to world matrix in opencv/colmap system, 4x4
• K: camera intrinsic parameter, 3x3
• width: image width
• height: image height
• FovX: fov x, calculated from K
• FovY: fov y, calculated from K
• image_path: relative path of image
• image_name: image name without .png/.jpg