H2OT hierarchically prunes video pose tokens to a few representative frames and recovers the full sequence, cutting FLOPs by 57-63% with near-identical MPJPE on four transformer pose estimators.
BlockGCN: Redefine topology awareness for skeleton-based action recognition,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
H$_{2}$OT: Hierarchical Hourglass Tokenizer for Efficient Video Pose Transformers
H2OT hierarchically prunes video pose tokens to a few representative frames and recovers the full sequence, cutting FLOPs by 57-63% with near-identical MPJPE on four transformer pose estimators.