L2M-GAN: Learning to Manipulate Latent Space Semantics for Facial Attribute Editing

Yang, Guo‐Yu; Fei, Nanyi; Ding, Mingyu; Liu, Guang‐Zhen; Lu, Zhiwu; Xiang, Tao

doi:10.1109/cvpr46437.2021.00297

Cited by 53 publications

(23 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In contrast, the second stream adopts an encoder for efficient image projection [9,11,[30][31][32][33][34][35][36][37][38][39], but due to the low-dimensional bottleneck layer (i.e., the innermost feature map with minimal spatial size), the encoder-decoder structure often faces the problem of inaccurate reconstruction [9,35,36]. Recent works often treat attribute manipulation as an image-to-image translation task [15,30,33,[35][36][37][38][39] and learn to synthesize an output image directly according to user input. Despite promising results, most image-to-image translations [15,[35][36][37][38][39][40] can only edit the predefined attributes before training, limiting their flexibility in the inference stage.…”

Section: Face Attribute Manipulationmentioning

confidence: 99%

“…Recent works often treat attribute manipulation as an image-to-image translation task [15,30,33,[35][36][37][38][39] and learn to synthesize an output image directly according to user input. Despite promising results, most image-to-image translations [15,[35][36][37][38][39][40] can only edit the predefined attributes before training, limiting their flexibility in the inference stage.…”

Section: Face Attribute Manipulationmentioning

confidence: 99%

“…Firstly, we present additional qualitative comparisons with the image-toimage translation models omitted in the manuscript, i.e., AttGAN [40], Star-GANV2 [36], and L2MGAN [15]. As can be seen, image-to-image translation models are often limited to the predefined attributes and cannot continuously manipulate the attribute.…”

Section: Appendix a Overviewmentioning

confidence: 99%

“…According to a fixed pre-trained GAN, recent works either map the latent vector to each attribute via subspace projection using attribute labels [1,17] or identify interpretable directions in the latent space of GAN in an unsupervised manner [6-8, 10, 18]. Although some attributes can be edited individually [1,6,8,15,16], the whole-face single-embedding manner limits the flexibility of editing facial components. Another recent line of approach resorts to geometryguided disentanglement [19][20][21][22][23], which requires additional geometric guidance in visual space as input, such as facial landmarks [19,20], segmentation mask [22] or 3D information [23], etc.…”

Section: Introductionmentioning

confidence: 99%

“…Figure D.25: Comparisons with three face attribute editing models. Since the three image-to-image translation methods are limited to their predefined attributes (smile for L2MGAN[15], male for StarGANV2[36], 13 attributes for AttGAN[40]), we only compare the performance on several common attributes.…”

mentioning

confidence: 99%

See 4 more Smart Citations

IA-FaceS: A Bidirectional Method for Semantic Face Editing

Huang¹,

Tu²,

Xu³

2022

Preprint

View full text Add to dashboard Cite

Semantic face editing has achieved substantial progress in recent years. Known as a growingly popular method, latent space manipulation performs face editing by changing the latent code of an input face to liberate users from painting skills.However, previous latent space manipulation methods usually encode an entire face into a single low-dimensional embedding, which constrains the reconstruction capacity and the control flexibility of facial components, such as eyes and nose. This paper proposes IA-FaceS as a bidirectional method for disentangled face attribute manipulation as well as flexible, controllable component editing without the need for segmentation masks or sketches in the original image. To strike a balance between the reconstruction capacity and the control flexibility, the encoder is designed as a multi-head structure to yield embeddings for reconstruction and control, respectively: a high-dimensional tensor with spatial properties for consistent reconstruction and four low-dimensional facial component embeddings for semantic face editing. A simple, fixed mask is devised over an intermediate latent feature tensor to segment component-specific regions and extract an embedding for each component separately through componentspecific sub-embedding layers. Manipulating the separate component embeddings can help achieve disentangled attribute manipulation and flexible control of facial components. To further disentangle the highly-correlated components,

show abstract

Section: Face Attribute Manipulationmentioning

confidence: 99%

Section: Face Attribute Manipulationmentioning

confidence: 99%

Section: Appendix a Overviewmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

mentioning

confidence: 99%

See 3 more Smart Citations

IA-FaceS: A Bidirectional Method for Semantic Face Editing

Huang¹,

Tu²,

Xu³

2022

Preprint

View full text Add to dashboard Cite

show abstract

Face attribute translation with multiple feature perceptual reconstruction assisted by style translator

Zhu,

Liang,

Liu

2024

Computer Animation & Virtual

View full text Add to dashboard Cite

Improving the accuracy and disentanglement of attribute translation, and maintaining the consistency of face identity have been hot topics in face attribute translation. Recent approaches employ attention mechanisms to enable attribute translation in facial images. However, due to the lack of accuracy in the extraction of style code, the attention mechanism alone is not precise enough for the translation of attributes. To tackle this, we introduce a style translator module, which partitions the style code into attribute‐related and unrelated components, enhancing latent space disentanglement for more accurate attribute manipulation. Additionally, many current methods use per‐pixel loss functions to preserve face identity. However, this can sacrifice crucial high‐level features and textures in the target image. To address this limitation, we propose a multiple‐perceptual reconstruction loss to better maintain image fidelity. Extensive qualitative and quantitative experiments in this article demonstrate significant improvements over state‐of‐the‐art methods, validating the effectiveness of our approach.

show abstract

CoPrGAN: Image-to-Image Translation via Content Preservation

Yu¹,

Zhou²

2022

Lecture Notes in Computer Science

View full text Add to dashboard Cite

L2M-GAN: Learning to Manipulate Latent Space Semantics for Facial Attribute Editing

Cited by 53 publications

References 29 publications

IA-FaceS: A Bidirectional Method for Semantic Face Editing

IA-FaceS: A Bidirectional Method for Semantic Face Editing

Face attribute translation with multiple feature perceptual reconstruction assisted by style translator

CoPrGAN: Image-to-Image Translation via Content Preservation

Contact Info

Product

Resources

About