TAPE: Temporal Attention-Based Probabilistic Human Pose and Shape Estimation

Reconstructing 3D human pose and shape from monocular videos is a well-studied but challenging problem. Common challenges include occlusions, the inherent ambiguities in 2D to mapping computational complexity of video processing. Existing methods ignore reconstruction provide single deterministic estimate for pose. In order address these issues, we present Temporal Attention based Probabilistic Estimation method (TAPE) that operates on an RGB video. More specifically, propose use neural network encode frames temporal features using attention-based network. Given features, output per-frame temporally-informed probability distribution Normalizing Flows. We show TAPE outperforms state-of-the-art standard benchmarks serves as effective video-based prior optimization-based estimation. Code available at: https: //github.com/nikosvasilik/TAPE