Linguistics-aware Masked Image Modeling for Self-supervised Scene Text Recognition

Benchmark Model Rank Results
scene-text-recognition-on-bctr-originalLMIM (extra data)–Scene LACC: 83.6Web LACC: 82.0Document LACC: 99.1…
scene-text-recognition-on-cute80LMIM (Union14M-L fine-tuning) (extra data)–Accuracy: 98.6
scene-text-recognition-on-iiit5kLMIM (Union14M-L fine-tuning) (extra data)–Accuracy: 98.5
scene-text-recognition-on-svtLMIM (Union14M-L fine-tuning) (extra data)–Accuracy: 97.7
scene-text-recognition-on-svtpLMIM (Union14M-L fine-tuning) (extra data)–Accuracy: 94.0
scene-text-recognition-on-union14m-benchmarkLMIM (20-epoch, ARD fine-tuning) (extra data)–WAICS (Curve): 90.6WAICS (General): 85.7WAICS (Multi-Oriented): 86.9
scene-text-recognition-on-union14m-benchmarkLMIM (20-epoch, Union14M-L fine-tuning) (extra data)–WAICS (Contextless): 85.5WAICS (Multi-Words): 87.9…