Localizing and Editing Knowledge In Text-to-Image Generative Models (ICLR2024) 今回はtext-to-imageの概念消去に戻ってDiffQuickFixという手法をみます. 論文は60ページ以上あり, 非常に多くの結果が提供されていますが主にmain paperの部分を見ていきます. 図や表はことわりのない限り論文からの引用です. 書籍情報 Samyadeep Basu and Nanxuan Zhao and Vlad I Morariu and Soheil Feizi and Varun Manjunatha. Localizing and Editing Knowledge In Text-to-Image Generative Models. The Twelfth International Con
![Stable Diffusionからの概念消去⑤:DiffQuickFix (論文)](https://cdn-ak-scissors.b.st-hatena.com/image/square/8b9103ae1aed30d78fac3f86c77eb9c4b2fd15d3/height=288;version=1;width=512/https%3A%2F%2Fres.cloudinary.com%2Fzenn%2Fimage%2Fupload%2Fs--p_LbzNDv--%2Fc_fit%252Cg_north_west%252Cl_text%3Anotosansjp-medium.otf_55%3AStable%252520Diffusion%2525E3%252581%25258B%2525E3%252582%252589%2525E3%252581%2525AE%2525E6%2525A6%252582%2525E5%2525BF%2525B5%2525E6%2525B6%252588%2525E5%25258E%2525BB%2525E2%252591%2525A4%2525EF%2525BC%25259ADiffQuickFix%252520%252528%2525E8%2525AB%252596%2525E6%252596%252587%252529%252Cw_1010%252Cx_90%252Cy_100%2Fg_south_west%252Cl_text%3Anotosansjp-medium.otf_37%3Afmuuly%252Cx_203%252Cy_121%2Fg_south_west%252Ch_90%252Cl_fetch%3AaHR0cHM6Ly9saDMuZ29vZ2xldXNlcmNvbnRlbnQuY29tL2EvQUNnOG9jSmRZdzZnbWtlMWlJMnY5R3dSRzNYcVB5MmxGdVQ1NVVzbDhvdUpKM0oyQ2dZPXM5Ni1j%252Cr_max%252Cw_90%252Cx_87%252Cy_95%2Fv1627283836%2Fdefault%2Fog-base-w1200-v2.png)