RZCR: Zero-shot Character Recognition via Radical-based Reasoning

Authors: Xiaolei Diao, Daqian Shi, Hao Tang, Qiang Shen, Yanzeng Li, Lei Wu, Hao Xu

IJCAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We validate our method on multiple datasets, and RZCR shows promising experimental results, especially on few-sample character datasets. 5 Experiments
Researcher Affiliation Academia 1College of Computer Science and Technology, Jilin University 2DISI, University of Trento 3CVL, ETH Zurich 4College of Software Engineering, Jilin University
Pseudocode Yes Algorithm 1 Weight Fusion-based Reasoning.
Open Source Code No The paper does not provide concrete access to source code for the described methodology. No links to repositories or explicit statements of code release are found.
Open Datasets Yes To evaluate our method on real-world character image sets which suffer from the few-sample problem, we introduce a new character image dataset called Oracle RC. We collect oracle rubbing images from [Wu, 2012] and normalize images by a denoising method [Shi et al., 2022a]. We also validate RZCR on handwritten Chinese datasets ICDAR2013 [Yin et al., 2013], HWDB1.1 [Liu et al., 2013], scene character dataset CTW [Yuan et al., 2019], and Korean dataset PE92 [KIM et al., 1996] for a comprehensive evaluation.
Dataset Splits No For all four datasets, we select 80% of the samples in each character category as the training set and the remaining as the test set.
Hardware Specification No The resolution of the input image is 416 416. We exploit data enhancement strategies, including translation, rotation, scaling, and background transformation. Parameters K=13 and M=3 are set in RIE. All experiments are conducted with Adadelta optimization where the hyperparameters are set to ρ=0.95 and ε=10 6.
Software Dependencies No All experiments are conducted with Adadelta optimization where the hyperparameters are set to ρ=0.95 and ε=10 6.
Experiment Setup Yes The resolution of the input image is 416 416. We exploit data enhancement strategies, including translation, rotation, scaling, and background transformation. Parameters K=13 and M=3 are set in RIE. All experiments are conducted with Adadelta optimization where the hyperparameters are set to ρ=0.95 and ε=10 6.