| name | liquidtad-efficient-method-temporal-action |
| description | LiquidTAD: Efficient temporal action detection via liquid neural dynamics. Replaces self-attention with parallelized ActionLiquid blocks for parameter-efficient video understanding. Use when building temporal action detection systems, deploying on edge devices, or replacing attention with liquid neural networks. |
| version | 1.0.0 |
| metadata | {"hermes":{"tags":["liquid-neural-network","temporal-action-detection","video-understanding","parameter-efficient","edge-deployment"],"source_paper":"LiquidTAD: An Efficient Method for Temporal Action Detection via Liquid Neural Dynamics (arXiv:2604.18274)"}} |
LiquidTAD: An Efficient Method for Temporal Action Detection via Liquid Neural Dynamics
Source
- arXiv: 2604.18274
- Authors: Zepeng Sun, Naichuan Zheng, Hailun Xia
- Published: 2026-04-20
- Categories: cs.CV
Abstract
Temporal Action Detection (TAD) in untrimmed videos is currently dominated by Transformer-based architectures. While high-performing, their quadratic computational complexity and substantial parameter redundancy limit deployment in resource-constrained environments. In this paper, we propose LiquidTAD, a novel parameter-efficient framework that replaces cumbersome self-attention layers with parallelized ActionLiquid blocks. Unlike traditional Liquid Neural Networks (LNNs) that suffer from sequen
Key Concepts
- Identify core methodology from the paper
- Understand the mathematical framework
- Note the evaluation protocol and datasets used
Implementation Notes
- Refer to the original paper for detailed equations and architecture
- Check for any available code repositories linked in the paper
- Consider computational requirements for reproduction
Activation Keywords
- liquid-neural-network, temporal-action-detection, video-understanding, parameter-efficient
