Self-supervised Character-to-Character Distillation for Text Recognition

ICCV 2023  ยท  Tongkun Guan, Wei Shen, Xue Yang, Qi Feng, Zekun Jiang, Xiaokang Yang ยท

When handling complicated text images (e.g., irregular structures, low resolution, heavy occlusion, and uneven illumination), existing supervised text recognition methods are data-hungry. Although these methods employ large-scale synthetic text images to reduce the dependence on annotated real images, the domain gap still limits the recognition performance. Therefore, exploring the robust text feature representations on unlabeled real images by self-supervised learning is a good solution. However, existing self-supervised text recognition methods conduct sequence-to-sequence representation learning by roughly splitting the visual features along the horizontal axis, which limits the flexibility of the augmentations, as large geometric-based augmentations may lead to sequence-to-sequence feature inconsistency. Motivated by this, we propose a novel self-supervised Character-to-Character Distillation method, CCD, which enables versatile augmentations to facilitate general text representation learning. Specifically, we delineate the character structures of unlabeled real images by designing a self-supervised character segmentation module. Following this, CCD easily enriches the diversity of local characters while keeping their pairwise alignment under flexible augmentations, using the transformation matrix between two augmented views from images. Experiments demonstrate that CCD achieves state-of-the-art results, with average performance gains of 1.38% in text recognition, 1.7% in text segmentation, 0.24 dB (PSNR) and 0.0321 (SSIM) in text super-resolution. Code is available at https://github.com/TongkunGuan/CCD.

PDF Abstract ICCV 2023 PDF ICCV 2023 Abstract
Task Dataset Model Metric Name Metric Value Global Rank Uses Extra
Training Data
Result Benchmark
Scene Text Recognition CUTE80 CCD-ViT-Base(ARD_2.8M) Accuracy 98.3 # 7
Scene Text Recognition CUTE80 CCD-ViT-Tiny(ARD_2.8M) Accuracy 95.8 # 10
Scene Text Recognition CUTE80 CCD-ViT-Small(ARD_2.8M) Accuracy 98.3 # 7
Scene Text Recognition HOST CCD-ViT-Base 1:1 Accuracy 77.3 # 3
Scene Text Recognition ICDAR2013 CCD-ViT-Base(ARD_2.8M) Accuracy 98.3 # 6
Scene Text Recognition ICDAR2013 CCD-ViT-Small(ARD_2.8M) Accuracy 98.3 # 6
Scene Text Recognition ICDAR2013 CCD-ViT-Tiny(ARD_2.8M) Accuracy 97.5 # 14
Scene Text Recognition IIIT5k CCD-ViT-Base(ARD_2.8M) Accuracy 98.0 # 7
Scene Text Recognition IIIT5k CCD-ViT-Tiny(ARD_2.8M) Accuracy 97.1 # 11
Scene Text Recognition IIIT5k CCD-ViT-Small(ARD_2.8M) Accuracy 98.0 # 7
self-supervised scene text recognition Scene Text Recognition Benchmarks CCD-ViT-Small Average Accuracy 84.9 # 1
Scene Text Recognition SVT CCD-ViT-Base(ARD_2.8M) Accuracy 97.8 # 8
Scene Text Recognition SVT CCD-ViT-Small(ARD_2.8M) Accuracy 96.4 # 9
Scene Text Recognition SVT CCD-ViT-Tiny(ARD_2.8M) Accuracy 96.0 # 10
Scene Text Recognition SVTP CCD-ViT-Tiny Accuracy 91.6 # 10
Scene Text Recognition SVTP CCD-ViT-Small Accuracy 92.7 # 9
Scene Text Recognition SVTP CCD-ViT-Base Accuracy 96.1 # 7
self-supervised scene text recognition TextSeg CCD-ViT-Small IoU (%) 84.8 # 1
self-supervised scene text recognition TextZoom CCD-ViT-Small Average PSNR (dB) 21.84 # 1
SSIM 0.7843 # 1
Scene Text Recognition WOST CCD-ViT-Base 1:1 Accuracy 86.0 # 3

Methods