[3D HPS] SynBody: Synthetic Dataset with Layered Human Models for 3D Human Perception and Modeling
Project page : https://maoxie.github.io/SynBody/
SynBody: Synthetic Dataset with Layered Human Models for 3D Human Perception and Modeling
Abstract Synthetic data has emerged as a promising source for 3D human research as it offers low-cost access to large-scale human datasets. To advance the diversity and annotation quality of human models, we introduce a new synthetic dataset, Synbody, with
maoxie.github.io
Paper : https://arxiv.org/abs/2303.17368
SynBody: Synthetic Dataset with Layered Human Models for 3D Human Perception and Modeling
Synthetic data has emerged as a promising source for 3D human research as it offers low-cost access to large-scale human datasets. To advance the diversity and annotation quality of human models, we introduce a new synthetic dataset, Synbody, with three ap
arxiv.org
이번에 읽을 논문은 2023년에 공개된 SenseTime을 비롯한 여러 연구 기관에서 발표한 "SynBody: Synthetic Dataset with Layered Human Models for 3D Human Perception and Modeling" 입니다. 이는 human mesh recovery 및 Human NeRF 분야를 연구하기 위해 생성된 large-scale synthetic dataset 입니다. 무려 10,000개의 human body model, 1000개의 action 및 다양한 viewpoint를 다루는 170만개 3D annotation이 있는 이미지로 구성됩니다.
다른 3d human body dataset과 비교한 자료는 아래와 같습니다.
SynBody dataset(S) 및 Real dataset(R)으로 학습하고 평가한 결과는 다음과 같습니다.
각 방법론에 따라 데이터세트를 적용하여 평가한 결과는 다음과 같습니다.
