Presentation Title

Learning an Invariant Domain for Image Watermarking

Presenter Information

Agnibh DasguptaFollow

Advisor Information

Xin Zhong

Location

MBSC Gallery Room 308 - G

Presentation Type

Oral Presentation

Start Date

4-3-2022 10:45 AM

End Date

4-3-2022 12:00 PM

Abstract

Self-supervised learning has gained popularity in recent years due to lack of annotated datasets and their cost. Learning effective spatially invariant representations of images in a self-supervised setting has a wide variety of applications in several downstream tasks such as image watermarking, image classification, object detection, semantic segmentation. Most of the popular approaches in this area of research involve contrastive learning, but with a focus on feature mining for classification. We propose a contrastive self-supervised training scheme and demonstrate the invariance in our domain using reconstruction. Additionally, we explore the application of our invariant domain on robust image watermarking.

This document is currently not available here.

COinS
 
Mar 4th, 10:45 AM Mar 4th, 12:00 PM

Learning an Invariant Domain for Image Watermarking

MBSC Gallery Room 308 - G

Self-supervised learning has gained popularity in recent years due to lack of annotated datasets and their cost. Learning effective spatially invariant representations of images in a self-supervised setting has a wide variety of applications in several downstream tasks such as image watermarking, image classification, object detection, semantic segmentation. Most of the popular approaches in this area of research involve contrastive learning, but with a focus on feature mining for classification. We propose a contrastive self-supervised training scheme and demonstrate the invariance in our domain using reconstruction. Additionally, we explore the application of our invariant domain on robust image watermarking.