Pan Zhou

Title: Pan Zhou - Innovator in Neural Network Based Scene Text Recognition

Introduction

Pan Zhou is a notable inventor based in Singapore, SG. He has made significant contributions to the field of artificial intelligence, particularly in the area of scene text recognition. His innovative work has led to the development of a unique neural network model that enhances the accuracy of text recognition from various scenes.

Latest Patents

Pan Zhou holds a patent for a system titled "Neural network based scene text recognition." This system utilizes a neural network-based model to perform scene text recognition with high accuracy. The architecture employs a double attention mechanism, which allows the model to effectively predict text from images. The model consists of a convolutional neural network component that generates visual features and an attention extractor neural network that calculates attention scores based on these features. By combining visual features and attention scores, the system produces mixed features that are input into a character recognizer component. This component further refines the attention score and recognizes characters accordingly. The training of the neural network model involves adjusting parameters to minimize a multi-class gradient harmonizing mechanism (GHM) loss, which varies based on the difficulty of the sample.

Career Highlights

Pan Zhou is currently employed at Salesforce, Inc., where he continues to innovate and contribute to advancements in technology. His work has garnered attention for its practical applications in enhancing text recognition capabilities.

Collaborations

Pan Zhou collaborates with talented individuals such as Peng Tang and Ran Xu, who contribute to the development and refinement of his innovative projects.

Conclusion

Pan Zhou's contributions to neural network-based scene text recognition exemplify the impact of innovation in technology. His work not only advances the field of artificial intelligence but also sets a foundation for future developments in text recognition systems.