Self-supervised Multi-view Stereo via Effective Co-Segmentation and Data-Augmentation
Authors: Hongbin Xu, Zhipeng Zhou, Yu Qiao, Wenxiong Kang, Qiuxia Wu3030-3038
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results on DTU dataset show that our proposed methods achieve the state-of-the-art performance among unsupervised methods, and even compete on par with supervised methods. Furthermore, extensive experiments on Tanks&Temples dataset demonstrate the effective generalization ability of the proposed method. |
| Researcher Affiliation | Academia | 1Shen Zhen Key Lab of Computer Vision and Pattern Recognition, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China 2Shanghai AI Lab, Shanghai, China 3South China University of Technology, Guangzhou, China |
| Pseudocode | No | The paper describes its method in detail through text and diagrams, but it does not include any explicit pseudocode or algorithm blocks. |
| Open Source Code | Yes | The code is released at: https://github.com/ Tough Stone X/Self-Supervised-MVS. |
| Open Datasets | Yes | Experimental results on DTU dataset show that our proposed methods achieve the state-of-the-art performance among unsupervised methods... (Aanæs et al. 2016). Furthermore, extensive experiments on Tanks&Temples dataset demonstrate the effective generalization ability of the proposed method. (Knapitsch et al. 2017) |
| Dataset Splits | Yes | During the training phase, we only use the training set of DTU without any ground truth depth maps. In default, the hyper-parameters during training and testing phase follow the same setting of Unsup MVS (Khot et al. 2019). |
| Hardware Specification | Yes | Our proposed JDACS is implemented in Pytorch and trained on 4 NVIDIA RTX 2080Ti GPUs. |
| Software Dependencies | No | The paper states that the method is 'implemented in Pytorch' and uses 'Adam optimizer' and 'pretrained VGG network', but it does not provide specific version numbers for any of these software components. |
| Experiment Setup | Yes | With a pattern of data-parallel, the batch size is set to 1 per GPU for JDACS and 4 per GPU for JDACS-MS, which consume no more than 10G memories in each GPU. We use Adam optimizer with a learning rate of 0.001 which decreases by 0.5 times for every two epochs. JDACS is trained for 10 epochs as MVSNet (Yao et al. 2018) and JDACS-MS is trained for 27 epochs as CVP-MVSNet(Yang et al. 2020). |