Object Discovery with a Copy-Pasting GAN (2019-05-27T00:00:00.000000Z)

TL;DR

A novel copy-pasting GAN framework is proposed, where the generator learns to discover an object in one image by compositing it into another image such that the discriminator cannot tell that the resulting image is fake.

Abstract

We tackle the problem of object discovery, where objects are segmented for a given input image, and the system is trained without using any direct supervision whatsoever. A novel copy-pasting GAN framework is proposed, where the generator learns to discover an object in one image by compositing it into another image such that the discriminator cannot tell that the resulting image is fake. After carefully addressing subtle issues, such as preventing the generator from `cheating', this game results in the generator learning to select objects, as copy-pasting objects is most likely to fool the discriminator. The system is shown to work well on four very different datasets, including large object appearance variations in challenging cluttered backgrounds.

Authors

Andrew Zisserman

62 papers

Relja Arandjelović

6 papers

TL;DR

Abstract

Authors

References83 items

Learning to Generate Synthetic Data via Compositing

Unsupervised Image Matching and Object Discovery as Optimization

Inserting Videos Into Videos

Multi-Object Representation Learning with Iterative Variational Inference

TF-Replicator: Distributed Machine Learning for Researchers

MONet: Unsupervised Scene Decomposition and Representation

Unsupervised Moving Object Detection via Contextual Information Separation

Context-aware Synthesis and Placement of Object Instances

Generate, Segment, and Refine: Towards Generic Manipulation Segmentation

Generate, Segment and Replace: Towards Generic Manipulation Segmentation

Large Scale GAN Training for High Fidelity Natural Image Synthesis

Learning to Read by Spelling: Towards Unsupervised Text Recognition

Semi-convolutional Operators for Instance Segmentation

Generative Adversarial Networks with Decoder-Encoder Output Noise

Learning Instance Segmentation by Interaction

Fighting Fake News: Image Splice Detection via Learned Self-Consistency

Synthesizing Programs for Images using Reinforced Adversarial Learning

Adversarial Network Compression

Patch-Based Image Inpainting with Generative Adversarial Networks

Learning to Segment via Cut-and-Paste

ST-GAN: Spatial Transformer Generative Adversarial Networks for Image Compositing

Path Aggregation Network for Instance Segmentation

Tempered Adversarial Networks

Low-Shot Learning from Imaginary Data

Unsupervised Cipher Cracking Using Discrete GANs

Recurrent Pixel Embedding for Instance Grouping

End-to-End Recovery of Human Shape and Pose

Unsupervised Machine Translation Using Monolingual Corpora Only

Progressive Growing of GANs for Improved Quality, Stability, and Variation

Word Translation Without Parallel Data

Image Splicing Localization using a Multi-task Fully Convolutional Network (MFCN)

Semantic Instance Segmentation with a Discriminative Loss Function

Cut, Paste and Learn: Surprisingly Easy Synthesis for Instance Detection

Adversarial Training for Unsupervised Bilingual Lexicon Induction

Show, Adapt and Tell: Adversarial Training of Cross-Domain Image Captioner

Semantic Instance Segmentation via Deep Metric Learning

Unpaired Image-to-Image Translation Using Cycle-Consistent Adversarial Networks

GP-GAN: Towards Realistic High-Resolution Image Blending

Mask R-CNN

Towards Diverse and Natural Image Descriptions via a Conditional GAN

Neural Expectation Maximization

Improved Texture Networks: Maximizing Quality and Diversity in Feed-Forward Stylization and Texture Synthesis

NIPS 2016 Tutorial: Generative Adversarial Networks

CLEVR: A Diagnostic Dataset for Compositional Language and Elementary Visual Reasoning

Image-to-Image Translation with Conditional Adversarial Networks

Associative Embedding: End-to-End Learning for Joint Detection and Grouping

Deconvolution and Checkerboard Artifacts

SeqGAN: Sequence Generative Adversarial Nets with Policy Gradient

Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network

Permutation invariant training of deep models for speaker-independent multi-talker speech separation

Generative Adversarial Imitation Learning

Adversarial Feature Learning

Precomputed Real-Time Texture Synthesis with Markovian Generative Adversarial Networks

Attend, Infer, Repeat: Fast Scene Understanding with Generative Models

Deep Residual Learning for Image Recognition

Rethinking the Inception Architecture for Computer Vision

Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks

Learning Dense Convolutional Embeddings for Semantic Segmentation

Learning to Segment Object Candidates

You Only Look Once: Unified, Real-Time Object Detection

U-Net: Convolutional Networks for Biomedical Image Segmentation

FlowNet: Learning Optical Flow with Convolutional Networks

Unsupervised object discovery and localization in the wild: Part-based matching with bottom-up region proposals

Adam: A Method for Stochastic Optimization

Unsupervised Domain Adaptation by Backpropagation

Fast Object Segmentation in Unconstrained Video

Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation

Unsupervised Joint Object Discovery and Segmentation in Internet Images

Video Object Segmentation through Spatially Accurate and Temporally Dense Extraction of Primary Object Regions

ImageNet classification with deep convolutional neural networks

“Clustering by Composition”—Unsupervised Discovery of Image Categories

Fast, automatic and fine-grained tampered JPEG image detection via DCT coefficient analysis

Detecting Digital Image Forgeries by Measuring Inconsistencies of Blocking Artifact

Unsupervised Learning of Categories from Sets of Partially Matching Image Features

Using Multiple Segmentations to Discover Objects and their Extent in Image Collections

Discovering objects and their location in images