IntegratedPIFu: Integrated Pixel Aligned Implicit Function for Single-View Human Reconstruction

Chan, Kennard Yanting; Lin, Guosheng; Zhao, Haiyu; Wang, Lin

doi:10.1007/978-3-031-20086-1_19

Cited by 17 publications

(10 citation statements)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The earliest sampling training scheme, called spatial sampling scheme, was introduced by the first pixel-aligned implicit model -PIFu (Saito et al 2019). Since then, subsequent pixel-aligned implicit models, such as PIFuHD (Saito et al 2020), PaMIR (Zheng et al 2021), ICON (Xiu et al 2022), S-PIFu (Chan et al 2022a), and more, simply continue to use the same scheme.…”

Section: )mentioning

confidence: 99%

“…IntegratedPIFu (Chan et al 2022b) is the first work to contribute a new sampling training scheme (i.e. Depth-Oriented Sampling or DOS) that differs materially from spatial sampling.…”

Section: )mentioning

confidence: 99%

“…After PIFu, other pixel-aligned implicit models have been proposed. These include PIFuHD (Saito et al 2020), Stere-oPIFu (Hong et al 2021), S-PIFu (Chan et al 2022a), Inte-gratedPIFu (Chan et al 2022b), and more.…”

Section: Single-view Human Reconstructionmentioning

confidence: 99%

“…In IntegratedPIFu (Chan et al 2022b), the authors identified problems with the spatial sampling scheme and proposed the Depth-Oriented Sampling (DOS) training scheme. They showed that the binary labels (as opposed to continuous labels) and unconstrained displacement of mesh sur- face points in spatial sampling scheme caused thin but obvious body features like the ears and fingers to be missing in the reconstructed meshes.…”

Section: Pifu and Integratedpifumentioning

confidence: 99%

“…An influential class of deep learning methods for singleimage clothed human reconstruction is pixel-aligned implicit models (Saito et al 2019(Saito et al , 2020Chan et al 2022b). These methods learn an implicit function that represents the surface of a human body.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Fine Structure-Aware Sampling: A New Sampling Training Scheme for Pixel-Aligned Implicit Models in Single-View Human Reconstruction

Chan,

Liu,

Lin

et al. 2024

AAAI

View full text Add to dashboard Cite

Pixel-aligned implicit models, such as PIFu, PIFuHD, and ICON, are used for single-view clothed human reconstruction. These models need to be trained using a sampling training scheme. Existing sampling training schemes either fail to capture thin surfaces (e.g. ears, fingers) or cause noisy artefacts in reconstructed meshes. To address these problems, we introduce Fine Structured-Aware Sampling (FSS), a new sampling training scheme to train pixel-aligned implicit models for single-view human reconstruction. FSS resolves the aforementioned problems by proactively adapting to the thickness and complexity of surfaces. In addition, unlike existing sampling training schemes, FSS shows how normals of sample points can be capitalized in the training process to improve results. Lastly, to further improve the training process, FSS proposes a mesh thickness loss signal for pixel-aligned implicit models. It becomes computationally feasible to introduce this loss once a slight reworking of the pixel-aligned implicit function framework is carried out. Our results show that our methods significantly outperform SOTA methods qualitatively and quantitatively. Our code is publicly available at https://github.com/kcyt/FSS.

show abstract

Section: )mentioning

confidence: 99%

“…IntegratedPIFu (Chan et al 2022b) is the first work to contribute a new sampling training scheme (i.e. Depth-Oriented Sampling or DOS) that differs materially from spatial sampling.…”

Section: )mentioning

confidence: 99%