Gradient-Based Graph Attention for Scene Text Image Super-resolution
Authors: Xiangyuan Zhu, Kehua Guo, Hui Fang, Rui Ding, Zheng Wu, Gerald Schaefer
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on the benchmark Text Zoom dataset convincingly demonstrate that our method supports excellent text recognition and outperforms the current state-of-the-art in STISR. |
| Researcher Affiliation | Academia | Xiangyuan Zhu1, Kehua Guo1*, Hui Fang2, Rui Ding1, Zheng Wu1, Gerald Schaefer2 1School of Computer Science and Engineering, Central South University, China 2Department of Computer Science, Loughborough University, U.K. |
| Pseudocode | No | No structured pseudocode or algorithm blocks were found in the paper. |
| Open Source Code | Yes | The source code is available at https://github.com/xyzhu1/TSAN. |
| Open Datasets | Yes | In our experiments, we use the benchmark Text Zoom dataset (Wang et al. 2020) to evaluate the performance of our proposed TSAN model and to compare it to the stateof-the-art. |
| Dataset Splits | No | The paper states: "Text Zoom is composed of two real text image super-resolution datasets, Real SR (Cai et al. 2019) and SRRAW (Zhang et al. 2019), in total containing 21,740 image pairs where the size of the low-resolution and high-resolution images are 16 64 and 32 128, respectively. Of these, 17,367 image pairs are used for training, and the remaining 4,373 pairs for testing". No explicit validation split information is provided. |
| Hardware Specification | Yes | All experiments are run, using Py Torch, on NVIDIA Ge Force RTX 2080Ti GPUs and an Intel i910940X processor @3.30 GHz. |
| Software Dependencies | No | The paper mentions "using Py Torch" but does not provide a specific version number for PyTorch or any other software dependencies. |
| Experiment Setup | Yes | We adopt the Adam optimiser (Kingma and Ba 2014) with β1 = 0.5 and β2 = 0.999 and train with a batch size of 16. We set the learning rate to 0.0001, and train the model for 300 epochs. ... Following our ablation studies, we set M = 8, λcha = 0.5, and use a patch size of 7 7. |