Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
DualNet: Robust Self-Supervised Stereo Matching with Pseudo-Label Supervision
Authors: Yun Wang, Jiahao Zheng, Chenghao Zhang, Zhanjie Zhang, Kunhong Li, Yongjian Zhang, Junjie Hu
AAAI 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We train and evaluate our model on four real-world datasets including KITTI 2012 (Geiger, Lenz, and Urtasun 2012), KITTI 2015 (Menze and Geiger 2015), Middlebury (Scharstein et al. 2014), ETH3D (Sch ops et al. 2017). ... Table 1: Comparative results achieved on the KITTI 2012 and KITTI 2015 benchmarks. ... Table 3: Ablation study on different losses for self-supervised training stage (step 1). |
| Researcher Affiliation | Academia | Yun Wang1, Jiahao Zheng1, Chenghao Zhang2, Zhanjie Zhang3, Kunhong Li4, Yongjian Zhang4 Junjie Hu5,6* 1 City University of Hong Kong 2 Chinese Academy of Sciences Institute of Automation, CASIA 3 Zhejiang University 4 Sun Yat-sen University 5 Shenzhen Institute of Artificial Intelligence and Robotics for Society 6 The Chinese University of Hong Kong (Shenzhen) |
| Pseudocode | No | The paper describes the methodology in narrative form, supplemented by diagrams like Figure 3, but does not include any explicit pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not contain any explicit statements regarding the public availability of source code or links to a code repository. |
| Open Datasets | Yes | We train and evaluate our model on four real-world datasets including KITTI 2012 (Geiger, Lenz, and Urtasun 2012), KITTI 2015 (Menze and Geiger 2015), Middlebury (Scharstein et al. 2014), ETH3D (Sch ops et al. 2017). |
| Dataset Splits | No | The paper states, 'We train and evaluate our model on four real-world datasets including KITTI 2012 (Geiger, Lenz, and Urtasun 2012), KITTI 2015 (Menze and Geiger 2015), Middlebury (Scharstein et al. 2014), ETH3D (Sch ops et al. 2017). Note that our training procedure does not require Ground Truth (GT) for these datasets. Instead, we only use GT for evaluation.' It also mentions 'using only a mixture of KITTI 12 & 15 image pairs (394 image pairs)' for training, but does not specify the exact training, validation, or test splits used for its own experimental setup, beyond using the GT for evaluation on the benchmark test sets. |
| Hardware Specification | Yes | We use 4 NVIDIA A6000 GPU and Pytorch with batch size 8 for all training experiments. |
| Software Dependencies | No | The paper mentions 'Pytorch' as a framework but does not specify its version number or any other software dependencies with version numbers. |
| Experiment Setup | Yes | For these two steps, the initial learning rate is set to 1 10 4 for the first 20 epochs and decreased to 1 10 5 for the remaining 80 epochs. We use 4 NVIDIA A6000 GPU and Pytorch with batch size 8 for all training experiments. The Adam W optimizer with β1= 0.9 and β2 = 0.999 is adopted. During training, 320 832 patches are randomly cropped. We use random cropping/color transformation and asymmetric occlusion for data augmentation. |