The Moving INfants In RGB-D (MINI-RGBD)

我们通过释放使用最近引入的皮肤多婴儿线性身体模型 (SMIL) 创建的rgb-d (MINI-RGBD) 数据集中的移动婴儿来填补这一空白。我们将真实的婴儿运动映射到具有逼真的形状和纹理的SMIL模型，并生成具有精确地面真相2D和3D关节位置的RGB和深度图像。我们使用最先进的方法评估我们的数据集，以进行RGB图像中的2D姿势估计以及深度图像中的3D姿势估计。

We address this gap by releasing the RGB-D (MINI-RGBD) dataset, which features moving infants and was constructed using the recently proposed Skin Multilinear Infant Body Model (SMIL). We map real infant movements onto the SMIL model with realistic shapes and textures, and generate RGB and depth images equipped with accurate ground-truth 2D and 3D joint positions. We evaluate our dataset using state-of-the-art methods for 2D pose estimation from RGB images and 3D pose estimation from depth images.