Deep Stereo Matching With Explicit Cost Aggregation Sub-Architecture
Authors: Lidong Yu, Yucheng Wang, Yuwei Wu, Yunde Jia
AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The comprehensive experiments on challenge datasets such as KITTI and Scene Flow show that our method outperforms the state-of-the-art methods. |
| Researcher Affiliation | Collaboration | 1Beijing Laboratory of Intelligent Information Technology, School of Computer Science, Beijing Institute of Technology, Beijing, 100081 2Kandao Australia Research Center Suite 3.05 1 Richardson Place North Ryde NSW 2113, Australia |
| Pseudocode | Yes | Algorithm 1: Deep Cost Aggregation |
| Open Source Code | No | The paper does not provide any explicit statement or link for the availability of its source code. |
| Open Datasets | Yes | We evaluate our method on three datasets, including Scene Flow (Mayer et al. 2016), KITTI2015 (Menze and Geiger 2015) and KITTI2012 (Geiger, Lenz, and Urtasun 2012). |
| Dataset Splits | Yes | Scene Flow is a synthetic data set for stereo matching which contains 35454 training and 4370 testing image pairs. ... The KITTI 2012 dataset contains 192 training and 195 testing images, and the KITTI 2015 dataset contains 200 training and 200 testing images. |
| Hardware Specification | Yes | We train the network on the Scene Flow dataset from a random initialization with shuffled orders. The training takes 23h after 300K iterations on a single NVIDIA 1080Ti GPU. |
| Software Dependencies | No | Our architecture is implemented by the Tensoflow (Abadi et al. 2016) with a standard RMSProp (Tieleman and Hinton 2012) and a constant learning rate of 0.0001. The paper mentions TensorFlow but does not specify its version number or versions for other key software components. |
| Experiment Setup | Yes | Our architecture is implemented by the Tensoflow (Abadi et al. 2016) with a standard RMSProp (Tieleman and Hinton 2012) and a constant learning rate of 0.0001. We train the network on the Scene Flow dataset from a random initialization with shuffled orders. The training takes 23h after 300K iterations on a single NVIDIA 1080Ti GPU. For the KITTI dataset, we fine-tune the model pre-trained on Scene Flow dataset with 70k iterations. Limited by the computation resource, we sub-sample all data by four times using the bilinear interpolation. |