Taihong Xiao

Title: Innovations of Taihong Xiao

Introduction

Taihong Xiao is a prominent inventor based in Merced, California. He has made significant contributions to the field of neural networks and image processing. With a total of two patents to his name, his work focuses on advancing the understanding of semantic correspondence through innovative training strategies.

Latest Patents

One of Taihong Xiao's latest patents is titled "Learning contrastive representation for semantic correspondence." This patent describes a multi-level contrastive training strategy for training a neural network using image pairs without additional labels. The approach aims to learn semantic correspondences at both the image level and the region or pixel level. The neural network is trained using contrasting image pairs that include different objects and corresponding image pairs that feature different views of the same object. The concept behind this training method is to pull corresponding image pairs closer together while pushing contrasting pairs apart. An image-level contrastive loss is computed from the neural network's outputs and is utilized to update the network's parameters through backpropagation. Additionally, the neural network is trained via pixel-level contrastive learning, which also relies solely on image pairs.

Another significant patent by Taihong Xiao is "Learnable cost volume for determining pixel correspondence." This method involves obtaining a first set of feature vectors associated with a first image and a second set of feature vectors associated with a second image. The process includes generating transformed feature vectors by applying a kernel matrix that defines an elliptical inner product space. A cost volume is then created by determining inner products between the transformed feature vectors and corresponding candidate feature vectors from the second set. This method ultimately facilitates the determination of pixel correspondence between the two images.

Career Highlights

Throughout his career, Taihong Xiao has worked with leading technology companies, including Nvidia Corporation and Google Inc. His experience in these organizations has allowed him to refine his skills and contribute to groundbreaking advancements in technology.

Collaborations

Taihong Xiao has collaborated with notable colleagues such as Sifei Liu and Shalini De Mello. These partnerships have further enriched his work and expanded the impact of his innovations.

Conclusion

Taihong Xiao's contributions to the field of neural networks and image processing are noteworthy. His innovative patents reflect a deep understanding of semantic correspondence and training strategies. Through his work, he continues to influence the landscape of technology and innovation.