# STARFlow: Spatial Temporal Feature Re-embedding with Attentive Learning for Real-world Scene Flow

arXiv: https://arxiv.org/abs/2403.07032 · HTML: https://mlsanity.com/paper/2403.07032/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 23 of its 34 references that are in the map.
- First posted on arXiv 11 Mar 2024.

## Builds on (most cited first, 23 of 23)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [ShapeNet: An Information-Rich 3D Model Repository](https://mlsanity.com/paper/1512.03012/) (2015)
- [PWC-Net: CNNs for Optical Flow Using Pyramid, Warping, and Cost Volume](https://mlsanity.com/paper/1709.02371/) (2017)
- [A Large Dataset to Train Convolutional Networks for Disparity, Optical Flow, and Scene Flow Estimation](https://mlsanity.com/paper/1512.02134/) (2015)
- [PointConv: Deep Convolutional Networks on 3D Point Clouds](https://mlsanity.com/paper/1811.07246/) (2018)
- [FlowNet3D: Learning Scene Flow in 3D Point Clouds](https://mlsanity.com/paper/1806.01411/) (2018)
- [FLOT: Scene Flow on Point Clouds Guided by Optimal Transport](https://mlsanity.com/paper/2007.11142/) (2020)
- [HPLFlowNet: Hierarchical Permutohedral Lattice FlowNet for Scene Flow Estimation on Large-scale Point Clouds](https://mlsanity.com/paper/1906.05332/) (2019)
- [PV-RAFT: Point-Voxel Correlation Fields for Scene Flow Estimation of Point Clouds](https://mlsanity.com/paper/2012.00987/) (2020)
- [FlowStep3D: Model Unrolling for Self-Supervised Scene Flow Estimation](https://mlsanity.com/paper/2011.10147/) (2020)
- [Weakly Supervised Learning of Rigid 3D Scene Flow](https://mlsanity.com/paper/2102.08945/) (2021)
- [FlowNet3D++: Geometric Losses For Deep Scene Flow Estimation](https://mlsanity.com/paper/1912.01438/) (2019)
- [Bi-PointFlowNet: Bidirectional Learning for Point Cloud Based Scene Flow Estimation](https://mlsanity.com/paper/2207.07522/) (2022)
- [Hierarchical Attention Learning of Scene Flow in 3D Point Clouds](https://mlsanity.com/paper/2010.05762/) (2020)
- [What Matters for 3D Scene Flow Network](https://mlsanity.com/paper/2207.09143/) (2022)
- [FESTA: Flow Estimation via Spatial-Temporal Attention for Scene Point Clouds](https://mlsanity.com/paper/2104.00798/) (2021)
- [Self-Supervised 3D Scene Flow Estimation Guided by Superpoints](https://mlsanity.com/paper/2305.02528/) (2023)
- [SCTN: Sparse Convolution-Transformer Network for Scene Flow Estimation](https://mlsanity.com/paper/2105.04447/) (2021)
- [milliFlow: Scene Flow Estimation on mmWave Radar Point Cloud for Human Motion Sensing](https://mlsanity.com/paper/2306.17010/) (2023)
- [Occlusion Guided Scene Flow Estimation on 3D Point Clouds](https://mlsanity.com/paper/2011.14880/) (2020)
- [Residual 3D Scene Flow Learning with Context-Aware Feature Extraction](https://mlsanity.com/paper/2109.04685/) (2021)
- [RMS-FlowNet: Efficient and Robust Multi-Scale Scene Flow Estimation for Large-Scale Point Clouds](https://mlsanity.com/paper/2204.00354/) (2022)

## Abstract (arXiv)

Scene flow prediction is a crucial underlying task in understanding dynamic scenes as it offers fundamental motion information. However, contemporary scene flow methods encounter three major challenges. Firstly, flow estimation solely based on local receptive fields lacks long-dependency matching of point pairs. To address this issue, we propose global attentive flow embedding to match all-to-all point pairs in both feature space and Euclidean space, providing global initialization before local refinement. Secondly, there are deformations existing in non-rigid objects after warping, which leads to variations in the spatiotemporal relation between the consecutive frames. For a more precise estimation of residual flow, a spatial temporal feature re-embedding module is devised to acquire the sequence features after deformation. Furthermore, previous methods perform poor generalization due to the significant domain gap between the synthesized and LiDAR-scanned datasets. We leverage novel domain adaptive losses to effectively bridge the gap of motion inference from synthetic to real-world. Experiments demonstrate that our approach achieves state-of-the-art performance across various datasets, with particularly outstanding results on real-world LiDAR-scanned datasets. Our code is available at https://github.com/O-VIGIA/StarFlow.