Adversarial Scene Editing: Automatic Object Removal from Weak\n Supervision

2018/06/05 by Rakshith Shetty, Mario Fritz, Shetty, Rakshith +3 · 4 citations
Computer Science · #Advanced Image Processing Techniques #Artificial Intelligence (cs.AI) #Computer Vision and Pattern Recognition (cs.CV) #Digital Media Forensic Detection #FOS: Computer and information sciences #Generative Adversarial Networks and Image Synthesis #Machine Learning (stat.ML)

paper · pdf · doi:10.48550/arxiv.1806.01911

openalex publication_date 2018/06/05 · openalex created_date 2022/09/03 · openalex updated_date 2026/07/28

Abstract

While great progress has been made recently in automatic image manipulation,\nit has been limited to object centric images like faces or structured scene\ndatasets. In this work, we take a step towards general scene-level image\nediting by developing an automatic interaction-free object removal model. Our\nmodel learns to find and remove objects from general scene images using\nimage-level labels and unpaired data in a generative adversarial network (GAN)\nframework. We achieve this with two key contributions: a two-stage editor\narchitecture consisting of a mask generator and image in-painter that\nco-operate to remove objects, and a novel GAN based prior for the mask\ngenerator that allows us to flexibly incorporate knowledge about object shapes.\nWe experimentally show on two datasets that our method effectively removes a\nwide variety of objects using weak supervision only\n

Cited by

• Privacy-Preserving Computer Vision for Industry: Three Case Studies in Human-Centric Manufacturing
• End-to-End Learning of Geometric Deformations of Feature Maps for Virtual Try-On
• Semantic Image Manipulation Using Scene Graphs
• A Closer Look at the Robustness of Vision-and-Language Pre-trained Models

Related

• EditGAN: High-Precision Semantic Image Editing
• Rethinking the editing of generative adversarial networks: a method to estimate editing vectors based on dimension reduction
• Navigating the GAN Parameter Space for Semantic Image Editing
• SEIGAN: Towards Compositional Image Generation by Simultaneously Learning to Segment, Enhance, and Inpaint
• Decorating Your Own Bedroom: Locally Controlling Image Generation with Generative Adversarial Networks