Toward Fine-Grained Facial Expression Manipulation

Ling, Jun; Han, Xue; Song, Li; Shu-hui, Yang; Xie, Rong; Gu, Xiao

doi:10.1007/978-3-030-58604-1_3

Cited by 27 publications

(34 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Based on GANimation, [17] adds a path for predicting an appearance flow to align the input image to the target expression. [5] changes the generator architecture of GANimation and uses relative AUs as input.…”

Section: Related Workmentioning

confidence: 99%

“…We choose 200000 pictures for training and 2000 pictures for testing. To extract the AU intensity vector, we use OpenFace [20] library to annotate intensities of 17 AUs [1,2,4,5,6,7,9,10,12,14,15,17,20,23,25,26,45]. Before training our model, all the facial pictures are resized to 128×128.…”

Section: Implementation Detailsmentioning

confidence: 99%

“…Before training our model, all the facial pictures are resized to 128×128. We compare our model with [4] and [5]. For convenience, we shortly write them as GANimation and fgGAN.…”

Section: Implementation Detailsmentioning

confidence: 99%

“…This mechanism will result in artifacts. To tackle these problems, [5] proposes a U-Net based generator.This method takes relative AUs which is the difference of source AUs and target AUs as input. Furthermore, it adopts several multi-scale feature fusion (MSF) modules in U-Net architecture for image transformation.…”

Section: Introductionmentioning

confidence: 99%

“…Fig.1shows an overview of our method. The model consists of a generator G and a discriminator D. We adopt the idea of[5] to set relative action unit vector and source image as input of the generator. In Fig.1, we conduct three generation steps.…”

mentioning

confidence: 99%

See 4 more Smart Citations

Attention Based Facial Expression Manipulation

Wang

Xiang

Liu

et al. 2021

2021 IEEE International Conference on Multimedia &Amp; Expo Workshops (ICMEW)

View full text Add to dashboard Cite

Facial expression manipulation has two objectives: 1) generating an image with target expression; 2) preserving the identity information of the original image as much as possible. Recently, Generative Adversarial Networks (GANs) have shown the abilities for fine-grained facial expression manipulation. However, current methods are still prone to generate images with poor quality. In this work, we propose a U-Net based generator with multi-attention gate for facial expression manipulation. The multi-level attention mechanism is helpful to manipulate relevant regions and preserve identity features, thus improving the editing ability. Furthermore, we adopt self-attention block to replace direct skip-connection to get long-range dependency in images. To suppress artifacts in generated images, we add a discriminator based loss function in the training process. Extensive experiments on both quantitative and qualitative evaluation show that our proposed method achieves better performance for facial expression manipulation.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Implementation Detailsmentioning

confidence: 99%

“…Before training our model, all the facial pictures are resized to 128×128. We compare our model with [4] and [5]. For convenience, we shortly write them as GANimation and fgGAN.…”

Section: Implementation Detailsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

mentioning

confidence: 99%

See 3 more Smart Citations