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].
Occlusion-Embedded Hybrid Transformer for Light Field Super-Resolution
Authors: Zeyu Xiao, Zhuoyuan Li, Wei Jia
AAAI 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on multiple datasets demonstrate OHT s superior performance. ... Extensive experiments on benchmark datasets show that our OHT can achieve superior performance. ... Quantitative and Qualitative Comparisons ... Ablation Study |
| Researcher Affiliation | Academia | Zeyu Xiao1, Zhuoyuan Li2, and Wei Jia3* 1National University of Singapore 2University of Science and Technology of China 3Hefei University of Technology |
| Pseudocode | No | The paper describes the methodology using textual descriptions and figures, but it does not contain a dedicated pseudocode block or algorithm section. |
| Open Source Code | No | The paper does not explicitly state that the source code for OHT or its implementation will be made publicly available. It mentions supplementary documents for details on OACCNet but not for the main proposed method's code. |
| Open Datasets | Yes | In line with prior works, we utilize the Basic LFSR benchmark (Wang 2023) for training and evaluating all methods at 2 and 4 scales. This benchmark includes five light field datasets: HCI-new (Honauer et al. 2016), HCI-old (Wanner, Meister, and Goldluecke 2013), EPFL (Rerabek and Ebrahimi 2016), INRIA (Le Pendu, Jiang, and Guillemot 2018), and STFGantry, comprising 144 training scenes and 23 test scenes with diverse contents and disparities. |
| Dataset Splits | Yes | This benchmark includes five light field datasets: HCI-new (Honauer et al. 2016), HCI-old (Wanner, Meister, and Goldluecke 2013), EPFL (Rerabek and Ebrahimi 2016), INRIA (Le Pendu, Jiang, and Guillemot 2018), and STFGantry, comprising 144 training scenes and 23 test scenes with diverse contents and disparities. ... In the training stage, we crop each view into 32 32 or 64 64 patches and perform 0.5 or 0.25 bicubic downsampling to generate LR patches for 2 and 4 SR, respectively. ... During training, we perform random horizontal flipping, vertical flipping, and 90-degree rotation to augment the data. |
| Hardware Specification | Yes | An NVIDIA A800 GPU is utilized for training. |
| Software Dependencies | No | All models are implemented using the Py Torch framework. The specific version of PyTorch is not provided. |
| Experiment Setup | Yes | We use the Xavier initialization algorithm and the Adam optimizer with β1 = 0.9 and β2 = 0.999. We set N = 6. The initial learning rate is set to 2.5 10 4 and decreases by a factor of 0.8 every 15 epochs. The batch size is set to 128. ... Specifically, the OHT-T and OHT-S models are trained from scratch for 75 epochs, while the OHT-B model is trained for 85 epochs. |