Stereoscopic Image Super-Resolution with Stereo Consistent Feature
Authors: Wonil Song, Sungil Choi, Somi Jeong, Kwanghoon Sohn12031-12038
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Finally, experimental results demonstrate the superiority of our method over state-of-the-art SR methods in terms of both quantitative metrics and qualitative visual quality while maintaining stereo-consistency between stereoscopic image pair. To evaluate our method, we conduct extensive experiments on Middlebury (Scharstein and Pal 2007), Flickr 1024 (Wang et al. 2019b), KITTI 2012 (Geiger, Lenz, and Urtasun 2012), and KITTI 2015 (Menze and Geiger 2015) datasets compared to the state-of-the-art SR methods. We also conduct an ablation study to analyze the contribution of our attention mechanism. |
| Researcher Affiliation | Academia | School of Electrical and Electronic Engineering, Yonsei University, Seoul, Korea {swonil92, csi6570, somijeong, khsohn}@yonsei.ac.kr |
| Pseudocode | No | The paper describes its method in text and with architectural diagrams (Figure 2 and Figure 3) but does not include explicit pseudocode or algorithm blocks. |
| Open Source Code | No | The paper references "Py Torch. https://github.com/pytorch/pytorch." which is a link to the PyTorch framework itself, not the specific source code implemented by the authors for their described methodology. There is no explicit statement about the release of their own code. |
| Open Datasets | Yes | We used the Middlebury (Scharstein and Pal 2007), Flickr 1024 (Wang et al. 2019b), and KITTI 2012 (Geiger, Lenz, and Urtasun 2012) and KITTI 2015 (Menze and Geiger 2015) dataset to train and evaluate our method. |
| Dataset Splits | Yes | To be specific, we divide the 60 Middlebury datasets into 30 pairs for training, 10 pairs for validation, and 20 pairs for evaluation. As provided the Flickr 1024 dataset, we used 800 pairs for trainining, 112 pairs for validation, and 112 pairs for evaluation. We select 40 pairs from KITTI 2012 and 2015 datasets, and they are used for only test. |
| Hardware Specification | Yes | Our network was implemented using Py Torch and trained on NVIDIA Ge Force GTX Titan X GPU. |
| Software Dependencies | No | The paper states: "Our network was implemented using Py Torch". While PyTorch is mentioned, no specific version number for PyTorch or any other software dependencies is provided, which is necessary for reproducibility. |
| Experiment Setup | Yes | The weights of networks are initialized by a Gaussian distribution with mean 0 and standard deviation 0.01, and the Adam optimizer (Kingma and Ba 2014) was employed for optimization, where β1 = 0.9, β2 = 0.999, and ϵ = 10 8. Additionally, for region-level SAM, we set k = 4. The initial learning rate is 10 4 and halved at every 30 epochs, and the batch size is 2. We set the parameters of the loss functions, such as λpa = 0.005, λstereo = 1. |