What Makes Fake Images Detectable? Understanding Properties that Generalize
The quality of image generation and manipulation is reaching impressive levels, making it increasingly difficult for a human to distinguish between what is real and what is fake. However, deep networks can still pick up on the subtle artifacts in these doctored images. We seek to understand what pro...
Main Authors: | , , |
---|---|
Other Authors: | |
Format: | Article |
Language: | English |
Published: |
Springer International Publishing
2021
|
Online Access: | https://hdl.handle.net/1721.1/129437.2 |
_version_ | 1811076132449026048 |
---|---|
author | Chai, Lucy Bau, David Isola, Phillip John |
author2 | Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory |
author_facet | Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory Chai, Lucy Bau, David Isola, Phillip John |
author_sort | Chai, Lucy |
collection | MIT |
description | The quality of image generation and manipulation is reaching impressive levels, making it increasingly difficult for a human to distinguish between what is real and what is fake. However, deep networks can still pick up on the subtle artifacts in these doctored images. We seek to understand what properties of fake images make them detectable and identify what generalizes across different model architectures, datasets, and variations in training. We use a patch-based classifier with limited receptive fields to visualize which regions of fake images are more easily detectable. We further show a technique to exaggerate these detectable properties and demonstrate that, even when the image generator is adversarially finetuned against a fake image classifier, it is still imperfect and leaves detectable artifacts in certain image patches. Code is available at https://github.com/chail/patch-forensics. |
first_indexed | 2024-09-23T10:16:43Z |
format | Article |
id | mit-1721.1/129437.2 |
institution | Massachusetts Institute of Technology |
language | English |
last_indexed | 2024-09-23T10:16:43Z |
publishDate | 2021 |
publisher | Springer International Publishing |
record_format | dspace |
spelling | mit-1721.1/129437.22021-09-09T18:01:58Z What Makes Fake Images Detectable? Understanding Properties that Generalize Chai, Lucy Bau, David Isola, Phillip John Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory The quality of image generation and manipulation is reaching impressive levels, making it increasingly difficult for a human to distinguish between what is real and what is fake. However, deep networks can still pick up on the subtle artifacts in these doctored images. We seek to understand what properties of fake images make them detectable and identify what generalizes across different model architectures, datasets, and variations in training. We use a patch-based classifier with limited receptive fields to visualize which regions of fake images are more easily detectable. We further show a technique to exaggerate these detectable properties and demonstrate that, even when the image generator is adversarially finetuned against a fake image classifier, it is still imperfect and leaves detectable artifacts in certain image patches. Code is available at https://github.com/chail/patch-forensics. 2021-09-09T18:01:57Z 2021-01-19T15:02:04Z 2021-09-09T18:01:57Z 2020-08 2020-08 2020-12-18T18:35:55Z Article http://purl.org/eprint/type/ConferencePaper 0302-9743 https://hdl.handle.net/1721.1/129437.2 Chai, Lucy et al. “What Makes Fake Images Detectable? Understanding Properties that Generalize.” ECCV 2020: 16th European Conference on Computer Vision, Lecture Notes in Computer Science, 12371. © 2020 The Author(s) en http://dx.doi.org/10.1007/978-3-030-58574-7_7 Lecture Notes in Computer Science Creative Commons Attribution-Noncommercial-Share Alike http://creativecommons.org/licenses/by-nc-sa/4.0/ application/octet-stream Springer International Publishing arXiv |
spellingShingle | Chai, Lucy Bau, David Isola, Phillip John What Makes Fake Images Detectable? Understanding Properties that Generalize |
title | What Makes Fake Images Detectable? Understanding Properties that Generalize |
title_full | What Makes Fake Images Detectable? Understanding Properties that Generalize |
title_fullStr | What Makes Fake Images Detectable? Understanding Properties that Generalize |
title_full_unstemmed | What Makes Fake Images Detectable? Understanding Properties that Generalize |
title_short | What Makes Fake Images Detectable? Understanding Properties that Generalize |
title_sort | what makes fake images detectable understanding properties that generalize |
url | https://hdl.handle.net/1721.1/129437.2 |
work_keys_str_mv | AT chailucy whatmakesfakeimagesdetectableunderstandingpropertiesthatgeneralize AT baudavid whatmakesfakeimagesdetectableunderstandingpropertiesthatgeneralize AT isolaphillipjohn whatmakesfakeimagesdetectableunderstandingpropertiesthatgeneralize |