CVPR_C3HOST

Forecasting of 3D Whole-body Human Poses with Grasping Objects

_________________________________________________________________________________________

Haitao Yan, Qiongjie Cui*, Jiexin Xie, Shijie Guo

Accepted by CVPR 2024

Qiongjie Cui is the Corresponding Author

[Paper] [Code] [Poster]

1. Video

2. Proposed Method

The proposed model involves two key components: internal-context encoding and cross-modal external interaction learning. The former mines the cross-context features to yield homogeneous features, while the latter aims to learn cross-modal external interaction between human and object. Moreover,we also calculate the distance information between the human and object, aligning object features with human joint features in the temporal domain. This alignment is crucial for capturing the nuanced dynamics of human-object interactions. The above components enable us to predict activities across all body and hand joints, ensuring high-precision whole-body human pose prediction, even during object grasping.

3. Results