Generating Key Postures of Bharatanatyam Adavus with Pose Estimation
Summary: arXiv:2603.29570v1 Announce Type: cross
The preservation of intangible cultural dances, deeply rooted in centuries of tradition, faces unique challenges in the digital age. Among these forms, Bharatanatyam—a classical Indian dance—is distinguished by its emphasis on codified adavus and precise key postures. The ability to accurately generate these postures is essential for maintaining both anatomical and stylistic integrity. Moreover, it facilitates effective documentation, analysis, and broader dissemination of the dance through digital platforms.
Introduction
Bharatanatyam is not only a dance but also a rich cultural expression that embodies a complex set of movements, gestures, and rhythms. The dance form is governed by strict structural and symbolic rules, which necessitates high fidelity in its representation. With the advent of technology, there is a growing need to harness digital tools to preserve this art form. This article introduces a novel approach utilizing a pose-aware generative framework integrated with a pose estimation module.
Methodology
Our framework incorporates keypoint-based loss and pose consistency constraints, acting as supervisory signals to ensure that the generated outputs maintain anatomical accuracy and stylistic integrity. We evaluated four configurations:
- Standard Conditional Generative Adversarial Network (cGAN)
- cGAN with pose supervision
- Conditional diffusion
- Conditional diffusion with pose supervision
Each model is conditioned on key posture class labels and optimized to preserve geometric structure. By integrating pose guidance in both the cGAN and conditional diffusion settings, we ensure that the generated poses align closely with the ground-truth keypoint structures, thereby enhancing cultural fidelity.
Results
The results of our experiments demonstrate that incorporating pose supervision significantly enhances the quality, realism, and authenticity of the generated Bharatanatyam postures. The models achieved a level of detail that respects both the anatomical movements and the intricate stylistic nuances of the dance form.
Implications for Cultural Preservation
This framework presents a scalable approach to the digital preservation and education of traditional dance forms. By enabling high-fidelity generation of Bharatanatyam postures, we are creating opportunities for wider access and understanding of this cultural heritage. The potential for digital dissemination paves the way for future generations to engage with and appreciate the art form.
Conclusion
In summary, the integration of pose estimation into generative models offers a promising pathway for the digital preservation of Bharatanatyam. Our approach not only maintains the intricate details of this classical dance but also ensures that the cultural precision is upheld in a digital context. Further research and development in this area could lead to enhanced educational tools and resources that benefit both practitioners and enthusiasts of this timeless art form.
For more information and access to the code, visit: Generating Key Postures of Bharatanatyam Adavus with Pose Estimation.