| Item type |
SIG Technical Reports(1) |
| 公開日 |
2023-02-27 |
| タイトル |
|
|
タイトル |
Digital Watermarks in Text-to-Image Generation Models |
| タイトル |
|
|
言語 |
en |
|
タイトル |
Digital Watermarks in Text-to-Image Generation Models |
| 言語 |
|
|
言語 |
eng |
| キーワード |
|
|
主題Scheme |
Other |
|
主題 |
プライバシー保護 |
| 資源タイプ |
|
|
資源タイプ識別子 |
http://purl.org/coar/resource_type/c_18gh |
|
資源タイプ |
technical report |
| 著者所属 |
|
|
|
Osaka University |
| 著者所属 |
|
|
|
Osaka University |
| 著者所属 |
|
|
|
Osaka University |
| 著者所属(英) |
|
|
|
en |
|
|
Osaka University |
| 著者所属(英) |
|
|
|
en |
|
|
Osaka University |
| 著者所属(英) |
|
|
|
en |
|
|
Osaka University |
| 著者名 |
Kunal, Paul
Yuntao, Wang
Atsuko, Miyaji
|
| 著者名(英) |
Kunal, Paul
Yuntao, Wang
Atsuko, Miyaji
|
| 論文抄録 |
|
|
内容記述タイプ |
Other |
|
内容記述 |
The rapid development of text-to-image generation models has raised concerns about privacy and security, as these models can scrape private images from the internet and use them in their training process. To address this issue, researchers have proposed using membership inference attacks (MIAs) to determine if an image has been used in the training process of a text-to-image generation model. However, the effectiveness of MIAs in these models is limited due to the unique nature of text-to-image creation. In this paper, we explore the limitations of conventional MIAs in text-to-image generation models and propose a solution to improve MIA performance through the use of digital watermarking. As far as we are aware, digital watermarking has not been previously suggested as a solution for MIA limitations in text-to-image generation models. Our contribution to this field includes a comprehensive examination of the issue and a practical solution to improve privacy and security in text-to-image generation models. |
| 論文抄録(英) |
|
|
内容記述タイプ |
Other |
|
内容記述 |
The rapid development of text-to-image generation models has raised concerns about privacy and security, as these models can scrape private images from the internet and use them in their training process. To address this issue, researchers have proposed using membership inference attacks (MIAs) to determine if an image has been used in the training process of a text-to-image generation model. However, the effectiveness of MIAs in these models is limited due to the unique nature of text-to-image creation. In this paper, we explore the limitations of conventional MIAs in text-to-image generation models and propose a solution to improve MIA performance through the use of digital watermarking. As far as we are aware, digital watermarking has not been previously suggested as a solution for MIA limitations in text-to-image generation models. Our contribution to this field includes a comprehensive examination of the issue and a practical solution to improve privacy and security in text-to-image generation models. |
| 書誌レコードID |
|
|
収録物識別子タイプ |
NCID |
|
収録物識別子 |
AN10116224 |
| 書誌情報 |
研究報告マルチメディア通信と分散処理（DPS）
巻 2023-DPS-194,
号 13,
p. 1-8,
発行日 2023-02-27
|
| ISSN |
|
|
収録物識別子タイプ |
ISSN |
|
収録物識別子 |
2188-8906 |
| Notice |
|
|
|
SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc. |
| 出版者 |
|
|
言語 |
ja |
|
出版者 |
情報処理学会 |
IPSJ-DPS23194013.pdf (6.3 MB)
