OAMaskFlow: Occlusion-Aware Motion Mask for Scene Flow
Authors: Xiongfeng Peng, Zhihua Liu, Weiming Li, Yamin Mao, Qiang Wang
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments show that our proposed OAMask Flow has reduced the EPE3D metric by 21.0% on the Flying Things3D dataset and decreased SF-all metric by 24.3% on the KITTI scene flow benchmark than the baseline method RAFT-3D. |
| Researcher Affiliation | Industry | Samsung R&D Institute China-Beijing, China EMAIL |
| Pseudocode | No | The paper describes the method using equations and textual explanations, but does not include any explicit pseudocode blocks or algorithms. |
| Open Source Code | No | The paper does not provide any explicit statements about releasing source code or links to a code repository. |
| Open Datasets | Yes | To validate our OAMask Flow method, we perform experiments on the synthetic Flying Things3D (Mayer et al. 2016) and the real-world KITTI (Menze and Geiger 2015) datasets. Furthermore, we extensively apply OAM mask into DROID-SLAM framework (Teed and Deng 2021a) and make experiments on the synthetic Tartan Air SLAM challenge (Wang et al. 2020a) dataset to validate our proposed OAM mask effectiveness. |
| Dataset Splits | Yes | Following Flow Net3D (Liu, Qi, and Guibas 2019), approximately 2000 test examples are sampled from the Flying Things3D test set for evaluation. |
| Hardware Specification | Yes | We train OAM-DROID-SLAM network on four Nvidia 3090 GPUs with two stages on Tartan Air (Wang et al. 2020a) dataset. |
| Software Dependencies | No | The paper uses existing frameworks and methods like RAFT-3D and DROID-SLAM but does not specify software dependencies with version numbers (e.g., Python, PyTorch, CUDA versions) for its own implementation. |
| Experiment Setup | Yes | We train our network for 200k iterations with a batch size of 4 and a crop size of [320, 720]. The initial learning rate is 2 10 4 and decays linearly during training. |