Iterative Token Evaluation and Refinement for Real-World Super-resolution
Authors: Chaofeng Chen, Shangchen Zhou, Liang Liao, Haoning Wu, Wenxiu Sun, Qiong Yan, Weisi Lin
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experimental results show that ITER is easy to train and performs well within just 8 iterative steps.We evaluate the performance of our model on multiple benchmarks that include real-world LQ images such as Real SR (Wang et al. 2021b), DReal SR (Wei et al. 2020), DPED-iphone (Ignatov et al. 2017), and Real SRSet (Zhang et al. 2021b).We performed a thorough analysis of various configurations of our model using a synthetic DIV2K validation test set. Firstly, we evaluated the effectiveness of refinement network in adding textures to the initial results Sl. Secondly, we assessed the necessity of the token evaluation block. |
| Researcher Affiliation | Collaboration | Chaofeng Chen1, Shangchen Zhou1, Liang Liao1, Haoning Wu1, Wenxiu Sun2, Qiong Yan2, Weisi Lin1 1 S-Lab, Nanyang Technological University 2 Sense Time Research |
| Pseudocode | Yes | Algorithm 1: Training of ITER, Algorithm 2: Adaptive Inference of ITER |
| Open Source Code | No | No explicit statement about open-source code release or a link to a repository was found. |
| Open Datasets | Yes | Our training dataset generation process follows that of Real-ESRGAN (Wang et al. 2021c), in which we obtain HQ images sourced from DIV2K (Agustsson and Timofte 2017), Flickr2K (Lim et al. 2017), and Outdoor Scene Training (Wang et al. 2018a). |
| Dataset Splits | Yes | Our training dataset generation process follows that of Real-ESRGAN (Wang et. al. 2021c)...and We performed a thorough analysis of various configurations of our model using a synthetic DIV2K validation test set. |
| Hardware Specification | Yes | All networks are implemented by Py Torch (Paszke et al. 2019) and trained for 400k iterations with 4 Tesla V100 GPUs. |
| Software Dependencies | No | All networks are implemented by Py Torch (Paszke et al. 2019) and trained for 400k iterations with 4 Tesla V100 GPUs. |
| Experiment Setup | Yes | The prominent Adam optimizer (Kingma and Ba 2014) is employed to optimize all three networks, with specific parameters of lr = 0.0001, β1 = 0.9, and β2 = 0.99. Each batch contains 16 HQ images of dimensions 256 256, paired with their corresponding LQ images. All networks are implemented by Py Torch (Paszke et al. 2019) and trained for 400k iterations with 4 Tesla V100 GPUs. |