Sequence-To-Sequence Domain Adaptation Network for Robust Text Image Recognition

Abstract

Training recognition models with synthetic images have achieved remarkable results in text recognition. However, recognizing text from real-world images still faces challenges due to the domain shift between synthetic and real-world text images. One of the strategies to eliminate the domain difference without manual annotation is unsupervised domain adaptation (UDA). Due to the characteristic of sequential labeling tasks, most popular UDA methods cannot be directly applied to text recognition. To tackle this problem, we proposed a UDA method with minimizing latent entropy on sequence-to-sequence attention-based models with classbalanced self-paced learning. Our experiments show that our proposed framework achieves better recognition results than the existing methods on most UDA text recognition benchmarks. All codes are publicly available 1 .

PGC-Net: A Light Weight Convolutional Sequence Network for Digital Pressure Gauge Calibration

“…On the other hand, Gao et al [31] proposed an end-to-end fully convolutional network with the stacked convolutional layers to capture the long-term dependencies among elements of scene text image. Besides, Zhang et al [32] develops a sequence-tosequence Domain Adaptation Network (SSDAN) that introduces a gated attention similarity unit to align the distribution of the source and target sequence data. Bartz et al [33] proposed a semi-supervised neural network for simultaneously scene text detection and recognition.…”

“…Bartz et al [33] proposed a semi-supervised neural network for simultaneously scene text detection and recognition. However, the method of [26], [32]- [35] adopted deep convolutional backbone to extract image feature and are too time-consuming for AGC.…”

GTC: Guided Training of CTC towards Efficient and Accurate Scene Text Recognition

“…Inspired by CTC, (Bai et al 2018) proposed a "Edit Probability" to optimize the training process, as missing or superfluity of characters may mislead CTC training. (Zhang et al 2019) also introduced a domain adaption method to varying length text recognition. The major approach for recent regular text recognition methods is still CTC-based, which enforces the alignment between feature sequence and labels.…”

Deep visual unsupervised domain adaptation for classification tasks: a survey

“…Chen et al [165] proposed a multinomial adversarial network (MAN) to address the text recognition problem by using the adversarial approach. In [166, 167], the encoder–decoder models are introduced for text recognition problem. Zhan et al [168] presented geometry‐aware domain adaptation network (GA‐DAN), which models the shift between domains in both geometry and appearance spaces, and converts images with different characteristics across domains.…”