Multi-task ConvNet for blind face inpainting with application to face verification

Shu, Zhang; He, Ran; Sun, Zhenan; Tan, Tieniu

doi:10.1109/icb.2016.7550058

Cited by 13 publications

(23 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this section, the proposed DF-GAN is evaluated on four MeshFace datasets, including three datasets under controlled conditions and one dataset in the wild. MeshFace is often corrupted by mesh-like occlusions that have random position, width and transparency [32,33]. Particularly, MeshFace completion requires both generating clean faces and improving verification performance.…”

Section: Methodsmentioning

confidence: 99%

“…Since large scale MeshFace images are difficult to obtain, we follow [32] to generate the MeshFace images by adding random patterns to clean faces. Firstly, the completely random binary patterns are synthesized with differently random magnitudes and phases.…”

Section: Baselines and Implementation Detailsmentioning

confidence: 99%

“…The parameter β controls the transparency of occlusions. Similar to [32], each clean face is used to synthesize 30 totally random MeshFaces on the AR, MultiPIE and Color FERET databases. Due to memory limitations, we only synthesize one random MeshFace for each clean face on CelebA and LFW databases.…”

Section: Baselines and Implementation Detailsmentioning

confidence: 99%

“…Our DF-GAN is able to disentangle and fuse clean faces and occlusions to realize the completion with unpaired data. Although [32,33] present their methods to recover clean faces from MeshFaces, they need to know the position of occlusions for training. Moreover, the paried clean faces are required for the recently proposed face completion methods [32,33,14,30].…”

Section: Baselines and Implementation Detailsmentioning

confidence: 99%

“…Finally, three domain-specific discriminators are imposed to compete with generators, which are trained by an adversarial strategy. Our DF-GAN is evaluated on Meshface verifica-tion problem [32,33], where MeshFace refers to a type of face images occluded by random mesh-like occlusions that have been widely used to protect private information from abusing [32,33]. Experimental results on four Meshface databases demonstrate the effectiveness of our proposed method for the face completion under structured occlusions.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Learning disentangling and fusing networks for face completion under structured occlusions

et al. 2020

Pattern Recognition

Self Cite

View full text Add to dashboard Cite

Face completion aims to generate semantically new pixels for missing facial components. It is a challenging generative task due to large variations of face appearance. This paper studies generative face completion under structured occlusions. We treat the face completion and corruption as disentangling and fusing processes of clean faces and occlusions, and propose a jointly disentangling and fusing Generative Adversarial Network (DF-GAN). First, three domains are constructed, corresponding to the distributions of occluded faces, clean faces and structured occlusions. The disentangling and fusing processes are formulated as the transformations between the three domains. Then the disentangling and fusing networks are built to learn the transformations from unpaired data, where the encoderdecoder structure is adopted and allows DF-GAN to simulate structure occlusions by modifying the latent representations. Finally, the disentangling and fusing processes are unified into a dual learning framework along with an adversarial strategy. The proposed method is evaluated on Meshface verification problem. Experimental results on four Meshface databases demonstrate the effectiveness of our proposed method for the face completion under structured occlusions.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Baselines and Implementation Detailsmentioning

confidence: 99%

Section: Baselines and Implementation Detailsmentioning

confidence: 99%