FaceFormer: Aggregating Global and Local Representation for Face Hallucination

Wang, Yuanzhi; Lü, Tao; Zhang, Yanduo; Wang, Zhongyuan; Jiang, Junjun; Xiong, Zixiang

doi:10.1109/tcsvt.2022.3224940

Cited by 22 publications

(4 citation statements)

References 51 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Wang et al. [51] introduced a novel multi‐path network, called FaceFormer, which incorporates the global representation provided by transformers with the local representation provided by CNNs. This allows for the preservation of facial structure consistency while also recovering local and global facial details.…”

Section: Related Workmentioning

confidence: 99%

“…Lu et al [50] proposed a FSR method using a split-attention network, consisting of two pathways for recovering facial structure and texture details, respectively. Wang et al [51] introduced a novel multi-path network, called FaceFormer, which incorporates the global representation provided by transformers with the local representation provided by CNNs. This allows for the preservation of facial structure consistency while also recovering local and global facial details.…”

Section: Face Super-resolution Networkmentioning

confidence: 99%

See 1 more Smart Citation

Efficient face image super‐resolution with convenient alternating projection network

Chen

et al. 2023

IET Signal Processing

View full text Add to dashboard Cite

The existing deep learning‐based face super‐resolution techniques can achieve satisfactory performance. However, these methods often incur large computational costs, and deeper networks generate redundant features. Some lightweight reconstruction networks also present limited representation ability because they ignore the entire contour and fine texture of the face for the sake of efficiency. Here, the authors propose a convenient alternating projection network (CAPN) for efficient face super‐resolution. First, the authors design a novel alternating projection block cascaded convolutional neural network to alternately achieve content consistency and learn detailed facial feature differences between super‐resolution and ground‐truth face images. Second, the self‐correction mechanism enabled the convolutional layer to capture faithful features that facilitate adaptive reconstruction. Moreover, a convenient connection operation can reduce the generation of redundant facial features while maintaining accurate reconstruction information. Extensive experiments demonstrated that the proposed CAPN can effectively reduce the computational cost while achieving competitive qualitative and quantitative results compared to state‐of‐the‐art super‐resolution methods.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Face Super-resolution Networkmentioning

confidence: 99%

Efficient face image super‐resolution with convenient alternating projection network

Chen

et al. 2023

IET Signal Processing

View full text Add to dashboard Cite

show abstract

“…On the other hand, Jiang et al [24] proposed the dual-path deep fusion network (DPDFN) for face image super-resolution without requiring an additional face prior, which learns the global facial shape and local facial components through two separate branches. Similarly, Wang et al [25] proposed the Face-Former model that combines the global representation of Transformers and local representation of CNNs to restore local facial details while maintaining the consistency of facial structure.…”

Section: Combined Global and Local Representation For Gementioning

confidence: 99%

“…Similarly, Wang et al. [25] proposed the Face‐Former model that combines the global representation of Transformers and local representation of CNNs to restore local facial details while maintaining the consistency of facial structure.…”

Section: Related Workmentioning

confidence: 99%

A novel gamut expansion method based on combined global–local mapping for sRGB‐to‐ProPhoto conversion

Liu

Sun

et al. 2023

IET Image Processing

View full text Add to dashboard Cite

With the development of display technology, more displays can cover the wider gamut, but most of the content they show is based on a small gamut. It is significant to employ gamut expansion (GE) to expand the small gamut images to a wider target gamut. Most of the existing GE methods only use global or local operations to realize the mapping from small gamut to wide gamut. However, the utilization of both global information and local feature is important for GE. In this article, the authors propose a combined global‐local gamut expansion network (G‐LGENet) for mapping the input standard RGB (sRGB) images to wider ProPhoto RGB space. In G‐LGENet, the global colour mapping module first extracts and fuses the global colour priors and learns the mapping of colour information for the corresponding pixels. And then, the local enhancement (LE) is designed to extract the local colour information between the corresponding pixel and neighbourhood pixels. The experimental results on a sRGB‐to‐ProPhoto dataset have demonstrated that the proposed G‐LGENet outperforms the other excellent GE methods qualitatively and visually.

show abstract