<p>Generative Adversarial Network (GAN) inversion have demonstrated excellent performance in image inpainting that aims to restore lost or damaged image texture using its unmasked content. Previous GAN inversion-based methods usually utilize well-trained GAN models as effective priors to generate the realistic regions for missing holes. Despite excellence, they ignore a hard constraint that the unmasked regions in the input and the output should be the same, resulting in a gap between GAN inversion and image inpainting and thus degrading the performance. Besides, existing GAN inversion approaches often consider a single modality of the input image, neglecting other auxiliary cues in images for improvements. Addressing these problems, we propose a novel GAN inversion approach, dubbed <i>MMInvertFill</i>, for image inpainting. MMInvertFill contains primarily a multimodal guided encoder with a pre-modulation and a GAN generator with <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11263_2025_2448_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="56" /> </InlineMediaObject> <EquationSource Format="TEX">$ \mathcal {F} \& \mathcal {W}^+$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi mathvariant="script">F</mi> <mo>&</mo> <msup> <mrow> <mi mathvariant="script">W</mi> </mrow> <mo>+</mo> </msup> </mrow> </math></EquationSource> </InlineEquation> latent space. Specifically, the multimodal encoder aims to enhance the multi-scale structures with additional semantic segmentation edge texture modalities through a gated mask-aware attention module. Afterwards, a pre-modulation is presented to encode these structures into style vectors. To mitigate issues of conspicuous color discrepancy and semantic inconsistency, we introduce the <InlineEquation ID="IEq2"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11263_2025_2448_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="56" /> </InlineMediaObject> <EquationSource Format="TEX">$ \mathcal {F} \& \mathcal {W}^+$</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi mathvariant="script">F</mi> <mo>&</mo> <msup> <mrow> <mi mathvariant="script">W</mi> </mrow> <mo>+</mo> </msup> </mrow> </math></EquationSource> </InlineEquation> latent space to bridge the gap between GAN inversion and image inpainting. Furthermore, in order to reconstruct faithful and photorealistic images, we devise a simple yet effective Soft-update Mean Latent module to capture more diversified in-domain patterns for generating high-fidelity textures for massive corruptions. In our extensive experiments on six challenging datasets, including CelebA-HQ, Places2, OST, CityScapes, MetFaces and Scenery, we show that our MMInvertFill qualitatively and quantitatively outperforms other state-of-the-arts and it supports the completion of out-of-domain images effectively. Our project webpage including code and results will be available at <a href="https://yeates.github.io/mm-invertfill">https://yeates.github.io/mm-invertfill</a>.</p>

High-Fidelity Image Inpainting with Multimodal Guided GAN Inversion

期刊论文
发表日期：

2025年5月16日

查看全文

Libo Zhang,
Yongsheng Yu,
Jiali Yao,
Heng Fan

摘要

Generative Adversarial Network (GAN) inversion have demonstrated excellent performance in image inpainting that aims to restore lost or damaged image texture using its unmasked content. Previous GAN inversion-based methods usually utilize well-trained GAN models as effective priors to generate the realistic regions for missing holes. Despite excellence, they ignore a hard constraint that the unmasked regions in the input and the output should be the same, resulting in a gap between GAN inversion and image inpainting and thus degrading the performance. Besides, existing GAN inversion approaches often consider a single modality of the input image, neglecting other auxiliary cues in images for improvements. Addressing these problems, we propose a novel GAN inversion approach, dubbed MMInvertFill, for image inpainting. MMInvertFill contains primarily a multimodal guided encoder with a pre-modulation and a GAN generator with $ \mathcal {F} \& \mathcal {W}^+$ $F & W^{+}$ latent space. Specifically, the multimodal encoder aims to enhance the multi-scale structures with additional semantic segmentation edge texture modalities through a gated mask-aware attention module. Afterwards, a pre-modulation is presented to encode these structures into style vectors. To mitigate issues of conspicuous color discrepancy and semantic inconsistency, we introduce the $ \mathcal {F} \& \mathcal {W}^+$ $F & W^{+}$ latent space to bridge the gap between GAN inversion and image inpainting. Furthermore, in order to reconstruct faithful and photorealistic images, we devise a simple yet effective Soft-update Mean Latent module to capture more diversified in-domain patterns for generating high-fidelity textures for massive corruptions. In our extensive experiments on six challenging datasets, including CelebA-HQ, Places2, OST, CityScapes, MetFaces and Scenery, we show that our MMInvertFill qualitatively and quantitatively outperforms other state-of-the-arts and it supports the completion of out-of-domain images effectively. Our project webpage including code and results will be available at https://yeates.github.io/mm-invertfill.