A continuous wavelet transform applied to per-joint velocities, followed by a lightweight multi-scale CNN, augments any skeleton backbone with explicit time-frequency dynamics and raises state-of-the-art gait recognition on CASIA-B.
Super- convergence: Very fast training of neural networks us- ing large learning rates
2 Pith papers cite this work. Polarity classification is still indexing.
2
Pith papers citing it
fields
cs.CV 2years
2026 2verdicts
UNVERDICTED 2representative citing papers
A multi-modal model with LMM semantic, ST visual, and PS audio branches enables simultaneous detection and fine-grained temporal localization of partial AI video forgeries, outperforming prior methods.
citing papers explorer
-
Explicit Time-Frequency Dynamics for Skeleton-Based Gait Recognition
A continuous wavelet transform applied to per-joint velocities, followed by a lightweight multi-scale CNN, augments any skeleton backbone with explicit time-frequency dynamics and raises state-of-the-art gait recognition on CASIA-B.
-
Towards multi-modal forgery representation learning for AI-generated video detection and localization
A multi-modal model with LMM semantic, ST visual, and PS audio branches enables simultaneous detection and fine-grained temporal localization of partial AI video forgeries, outperforming prior methods.