Closed Lizhinwafu closed 3 weeks ago
Thanks for your interest!
EmbodiedSAM is an online segmentation model, which requires the raw sensor input, i.e., the posed RGB-D video. However, S3DIS seems to only provide complete point clouds after 3D reconstruction, which is an offline dataset.
To run EmbodiedSAM, you should prepare a posed RGB-D sequence with: RGB frames, Depth frames, poses and camera intrinsics. You can follow our data preparation for ScanNet to organize your own data.
Thanks.
Can this code train the s3dis dataset? How can I make my own dataset?