International Conference on Computer and Knowledge Engineering

Home / 15th International Conference on Computer and Knowledge Engineering

Graph-Cut-Based Semantic Optimization for Temporal Action Segmentation

Authors :

Mohanna Ansari¹ Ehsan Fazl-Ersi²

1- Department of Computer Engineering, Ferdowsi University of Mashhad, Iran 2- Department of Computer Engineering, Ferdowsi University of Mashhad, Iran

Keywords :

Temporal action segmentation،Energy minimization،Graph-cut،Smooth Action Transition

Abstract :

Temporal action segmentation in untrimmed videos is critical for understanding human activities in applications such as robotics, surveillance, and human-computer interaction. While existing methods based on temporal convolutional networks (TCNs) and transformers effectively capture temporal dependencies and refine features, they often lack explicit mechanisms to enforce semantic consistency between action labels, leading to fragmented predictions. To address this limitation, we propose a novel framework that formulates temporal action segmentation as an energy minimization problem combining data fidelity and smoothness costs. Data costs are derived from a diffusion-based generative model (DiffAct) to capture action probabilities, while smoothness costs enforce semantic coherence by modeling valid transitions between action labels. We leverage graph-cut optimization to efficiently minimize the energy function. Experiments on the GTEA dataset demonstrate that our method, GBSO, achieves superior segmentation accuracy and temporal consistency compared to state-of-the-art approaches, improving boundary alignment and ensuring smoother semantic transitions. These results highlight the effectiveness of integrating semantic smoothness constraints into data-driven action segmentation frameworks.