EmbodiedSAM: Online Segment Any 3D Thing in Real Time
EmbodiedSAM is an efficient framework that leverages vision foundation models for online, real-time, fine-grained, generalized and open-vocabulary 3D instance segmentation.
Abstract
Embodied tasks require the agent to fully understand 3D scenes simultaneously with its exploration, so an online, real-time, fine-grained and highly-generalized 3D perception model is desperately needed. Since high-quality 3D data is limited, directly training such a model in 3D is almost infeasible. Meanwhile, vision foundation models (VFM) has revolutionized the field of 2D computer vision with superior performance, which makes the use of VFM to assist embodied 3D perception a promising direction. However, most existing VFM-assisted 3D perception methods are either offline or too slow that cannot be applied in practical embodied tasks. In this paper, we aim to leverage Segment Anything Model (SAM) for real-time 3D instance segmentation in an online setting. This is a challenging problem since future frames are not available in the input streaming RGB-D video, and an instance may be observed in several frames so object matching between frames is required. To address these challenges, we first propose a geometric-aware query lifting module to represent the 2D masks generated by SAM by 3D-aware queries, which is then iteratively refined by a dual-level query decoder. In this way, the 2D masks are transferred to fine-grained shapes on 3D point clouds. Benefit from the query representation for 3D masks, we can compute the similarity matrix between the 3D masks from different views by efficient matrix operation, which enables real-time inference. Experiments on ScanNet, ScanNet200, SceneNN and 3RScan show our method achieves leading performance even compared with offline methods. Our method also demonstrates great generalization ability in several zero-shot dataset transferring experiments and show great potential in open-vocabulary and data-efficient setting. Code and demo are available at https://xuxw98.github.io/ESAM/, with only one RTX 3090 GPU required for training and evaluation.
The official code is publicly release in this repo.
Citation
@article{xu2024esam,
title={EmbodiedSAM: Online Segment Any 3D Thing in Real Time},
author={Xiuwei Xu and Huangxing Chen and Linqing Zhao and Ziwei Wang and Jie Zhou and Jiwen Lu},
journal={arXiv preprint arXiv:2408.11811},
year={2024}
}
Main Results
We provide the checkpoints for quick reproduction of the results reported in the paper.
Class-agnostic 3D instance segmentation results on ScanNet200 dataset:
| Method | Type | VFM | AP | AP@50 | AP@25 | Speed(ms) | Downloads |
|---|---|---|---|---|---|---|---|
| SAMPro3D | Offline | SAM | 18.0 | 32.8 | 56.1 | -- | -- |
| SAI3D | Offline | SemanticSAM | 30.8 | 50.5 | 70.6 | -- | -- |
| SAM3D | Online | SAM | 20.6 | 35.7 | 55.5 | 1369+1518 | -- |
| ESAM | Online | SAM | 42.2 | 63.7 | 79.6 | 1369+80 | model |
| ESAM-E | Online | FastSAM | 43.4 | 65.4 | 80.9 | 20+80 | model |
Dataset transfer results from ScanNet200 to SceneNN and 3RScan:
| Method | Type | ScanNet200-->SceneNN | ScanNet200-->3RScan | ||||
|---|---|---|---|---|---|---|---|
| AP | AP@50 | AP@25 | AP | AP@50 | AP@25 | ||
| SAMPro3D | Offline | 12.6 | 25.8 | 53.2 | 3.9 | 8.0 | 21.0 |
| SAI3D | Offline | 18.6 | 34.7 | 65.7 | 5.4 | 11.8 | 27.4 |
| SAM3D | Online | 15.1 | 30.0 | 51.8 | 6.2 | 13.0 | 33.9 |
| ESAM | Online | 28.8 | 52.2 | 69.3 | 14.1 | 31.2 | 59.6 |
| ESAM-E | Online | 28.6 | 50.4 | 71.0 | 13.9 | 29.4 | 58.8 |
3D instance segmentation results on ScanNet dataset:
| Method | Type | ScanNet | SceneNN | FPS | Download | ||||
|---|---|---|---|---|---|---|---|---|---|
| AP | AP@50 | AP@25 | AP | AP@50 | AP@25 | ||||
| TD3D | offline | 46.2 | 71.1 | 81.3 | -- | -- | -- | -- | -- |
| Oneformer3D | offline | 59.3 | 78.8 | 86.7 | -- | -- | -- | -- | -- |
| INS-Conv | online | -- | 57.4 | -- | -- | -- | -- | -- | -- |
| TD3D-MA | online | 39.0 | 60.5 | 71.3 | 26.0 | 42.8 | 59.2 | 3.5 | -- |
| ESAM-E | online | 41.6 | 60.1 | 75.6 | 27.5 | 48.7 | 64.6 | 10 | model |
| ESAM-E+FF | online | 42.6 | 61.9 | 77.1 | 33.3 | 53.6 | 62.5 | 9.8 | model |
Open-Vocabulary 3D instance segmentation results on ScanNet200 dataset:
| Method | AP | AP@50 | AP@25 |
|---|---|---|---|
| SAI3D | 9.6 | 14.7 | 19.0 |
| ESAM | 13.7 | 19.2 | 23.9 |