Scalable 3D Panoptic Segmentation As Superpoint Graph Clustering
We introduce a highly efficient method for panoptic segmentation of large 3D point clouds by redefining this task as a scalable graph clustering problem. This approach can be trained using only local auxiliary tasks, thereby eliminating the resource-intensive instance-matching step during training. Moreover, our formulation can easily be adapted to the superpoint paradigm, further increasing its efficiency. This allows our model to process scenes with millions of points and thousands of objects in a single inference. Our method, called SuperCluster, achieves a new state-of-the-art panoptic segmentation performance for two indoor scanning datasets: $50.1$ PQ ($+7.8$) for S3DIS Area~5, and $58.7$ PQ ($+25.2$) for ScanNetV2. We also set the first state-of-the-art for two large-scale mobile mapping benchmarks: KITTI-360 and DALES. With only $209$k parameters, our model is over $30$ times smaller than the best-competing method and trains up to $15$ times faster. Our code and pretrained models are available at https://github.com/drprojects/superpoint_transformer.
PDF AbstractTask | Dataset | Model | Metric Name | Metric Value | Global Rank | Benchmark |
---|---|---|---|---|---|---|
Panoptic Segmentation | DALES | SuperCluster | PQ | 61.2 | # 1 | |
RQ | 68.6 | # 1 | ||||
SQ | 87.1 | # 1 | ||||
Params (M) | 0.21 | # 1 | ||||
3D Semantic Segmentation | DALES | SuperCluster | mIoU | 77.3 | # 3 | |
Model size | 210M | # 8 | ||||
Panoptic Segmentation | KITTI-360 | SuperCluster | PQ | 48.3 | # 1 | |
RQ | 58.4 | # 1 | ||||
SQ | 75.1 | # 1 | ||||
Params (M) | 0.79 | # 1 | ||||
3D Semantic Segmentation | KITTI-360 | SuperCluster | miou Val | 62.1 | # 2 | |
Model size | 790K | # 1 | ||||
Panoptic Segmentation | S3DIS | SuperCluster | PQ | 55.9 | # 1 | |
RQ | 66.3 | # 1 | ||||
SQ | 83.8 | # 1 | ||||
PQ (with stuff) | 62.7 | # 1 | ||||
RQ (with stuff) | 73.2 | # 1 | ||||
SQ (with stuff) | 84.8 | # 1 | ||||
Params (M) | 0.21 | # 1 | ||||
Semantic Segmentation | S3DIS | SuperCluster | Mean IoU | 75.3 | # 10 | |
Number of params | 0.21M | # 35 | ||||
Params (M) | 0.21 | # 18 | ||||
Semantic Segmentation | S3DIS Area5 | SuperCluster | mIoU | 68.1 | # 30 | |
Number of params | 0.21 | # 1 | ||||
Panoptic Segmentation | S3DIS Area5 | SuperCluster | PQ | 50.1 | # 1 | |
RQ | 60.1 | # 1 | ||||
SQ | 76.6 | # 1 | ||||
PQ (with stuff) | 58.4 | # 1 | ||||
RQ (with stuff) | 68.4 | # 1 | ||||
SQ (with stuff) | 77.8 | # 1 | ||||
Params (M) | 0.21 | # 1 | ||||
Panoptic Segmentation | ScanNet | SuperCluster | PQ | 58.7 | # 2 | |
PQ_th | 69.1 | # 2 | ||||
PQ_st | 84.1 | # 2 | ||||
Panoptic Segmentation | ScanNetV2 | SuperCluster | PQ | 58.7 | # 2 | |
SQ | 84.1 | # 1 | ||||
RQ | 69.1 | # 1 | ||||
Params (M) | 1 | # 1 |