Paper Image

Detecting Real vs. Synthetic Images

Published on:

21 September 2023

Primary Category:

Computer Vision and Pattern Recognition

Paper Authors:

Davide Cozzolino,

Koki Nagano,

Lucas Thomaz,

Angshul Majumdar,

Luisa Verdoliva

Bullets

Key Details

2022 IEEE VIP Cup challenged detecting real vs synthetic images

New AI can generate highly realistic fake visual content

Detecting fakes is crucial for trust in online media

Competition tested generalization and robustness

Top solutions used ensembles trained on diverse data

Difficult to generalize to new models unseen in training

Hard to set detection thresholds for open sets

AI generated summary

Detecting Real vs. Synthetic Images

This paper summarizes the 2022 IEEE Video and Image Processing Cup student competition, which challenged teams to develop systems for distinguishing real photos from AI-generated synthetic images. With new generative models producing highly realistic fake visual content, detecting synthesized media is crucial for maintaining trust online. The competition tested generalization on new architectures and robustness to compression. The top solutions used ensembles of deep networks trained on diverse augmented data. Key findings were the difficulty of generalizing to novel models and properly setting detection thresholds in open sets.

Answers from this paper

Comments

No comments yet, be the first to start the conversation...

Sign up to comment on this paper

Sign Up