Detail-Preserving Transformer for Light Field Image Super-resolution

Authors: Shunzhou Wang, Tianfei Zhou, Yao Lu, Huijun Di2522-2530

AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Evaluations are conducted on a number of light field datasets, including real-world scenes and synthetic data. The proposed method achieves superior performance comparing with other state-of-the-art schemes.
Researcher Affiliation Academia 1 Beijing Key Laboratory of Intelligent Information Technology, School of Computer Science and Technology, Beijing Institute of Technology, China 2 Computer Vision Laboratory, ETH Zurich, Switzerland
Pseudocode No The paper does not contain any pseudocode or algorithm blocks.
Open Source Code Yes Our code is publicly available at: https://github.com/BITszwang/DPT.
Open Datasets Yes We conduct extensive experiments on five popular LFSR benchmarks, i.e., EPFL (Rerabek and Ebrahimi 2016), HCInew (Honauer et al. 2016), HCIold (Wanner, Meister, and Goldluecke 2013), INRIA (Le Pendu, Jiang, and Guillemot 2018), and STFgantry (Vaish and Adams 2008).
Dataset Splits No The paper mentions 'training stage' and 'testing dataset' but does not specify explicit train/validation/test splits with percentages or counts for reproducibility. It implies a test set but no detailed split information for validation.
Hardware Specification Yes All experiments are carried out on a single Tesla V100 GPU card.
Software Dependencies No The paper mentions using the 'Adam optimizer' and 'ℓ1 loss' but does not specify any software names or versions (e.g., PyTorch 1.9, CUDA 11.1).
Experiment Setup Yes The ℓ1 loss is used to optimize our network. We use the Adam optimizer to train our network, with a batch size of 8. The initial learning rate is set to 2 10 4 and it will be halved every 15 epochs. We train the network for 75 epochs in total.