What question did this study set out to answer?

Develop a unified, end-to-end trainable neural network framework integrating feature extraction, sequence modeling, and transcription for image-based sequence and scene text recognition.

December 29, 2016IEEE Transactions on Pattern Analysis and Machine Intelligence

An End-to-End Trainable Neural Network for Image-Based Sequence Recognition and Its Application to Scene Text Recognition

View Full Paper

Ask AI

Bookmark

Authors

BSBaoguang ShiChinese Academy of Sciences XBXiang BaiKaramay Central Hospital of Xinjiang CYCong YaoJilin Normal University

Discussion

Member takes

Implication

Benchmarking study demonstrates superior accuracy in image-based sequence recognition across standard text datasets, highlighting the utility of unified end-to-end neural networks.

Key Points

Develop a unified, end-to-end trainable neural network framework integrating feature extraction, sequence modeling, and transcription for image-based sequence and scene text recognition.
Designed a neural network architecture combining visual feature extraction, sequence modeling, and transcription into a unified pipeline operating without character segmentation.
Evaluated the model across standard scene text benchmarks, including IIIT-5K, Street View Text, and ICDAR datasets, in both lexicon-free and lexicon-based settings.
Tested generalizability on an image-based music score recognition benchmark.
Outperformed existing state-of-the-art methods across standard benchmark datasets (IIIT-5K, Street View Text, and ICDAR) without requiring horizontal scale normalization.
Produced an effective, significantly smaller model architecture that handles arbitrary sequence lengths across both lexicon-based and lexicon-free tasks.
Demonstrated high performance on image-based music score recognition, confirming cross-domain applicability for visual sequence tasks.

Cite This Study

Shi et al. (2016) studied this question.

synapsesocial.com/papers/69d6f3475413bc3de5ab3128 https://doi.org/10.1109/tpami.2016.2646371

View Full Paper

Ask AI

Bookmark

Implication

Benchmarking study demonstrates superior accuracy in image-based sequence recognition across standard text datasets, highlighting the utility of unified end-to-end neural networks.

An End-to-End Trainable Neural Network for Image-Based Sequence Recognition and Its Application to Scene Text Recognition

Discussion

Member takes

Implication

Key Points

Cite This Study

An End-to-End Trainable Neural Network for Image-Based Sequence Recognition and Its Application to Scene Text Recognition

Discussion

Member takes

Implication

Key Points

Cite This Study

Also Consider

Also Consider