Sayan Nag

Title: Sayan Nag and Weakly-Supervised Referring Expression Segmentation.

Introduction

Sayan Nag is an inventor based in Toronto, Canada. His patent work concerns computer vision and methods for training systems to identify objects in images using text prompts and image data.

Latest Patents

Nag is associated with 1 patent, titled “Weakly-supervised referring expression segmentation.” The invention describes systems and methods that use a fused feature extractor to generate inferred masks from digital images and text prompts. The system selects masks that meet a validity threshold and combines them with ground-truth masks to create an augmented training set. It then uses object-mask predictions and ground-truth and pseudo-loss measures to adjust the feature extractor’s parameters.

Career Highlights

Nag’s listed patent is associated with Adobe, Inc. It reflects work on machine-learning techniques for image segmentation, particularly approaches that can use inferred labels alongside ground-truth data.

Collaborations

The provided information lists Koustava Goswami and Srikrishna Karanam as Nag’s coworkers. Their names are associated with his professional context, though specific roles in the patent are not stated.

Conclusion

Sayan Nag’s patent highlights an approach to improving referring-expression segmentation by combining inferred image masks with ground-truth training data.