Generating Key Postures of Bharatanatyam Adavus with Pose Estimation

Jagadish Kashinath Kamble; Jayanta Mukhopadhyay; Debaditya Roy; Partha Pratim Das

cs.CVMar 31, 2026

Generating Key Postures of Bharatanatyam Adavus with Pose Estimation

Authors: Jagadish Kashinath Kamble, Jayanta Mukhopadhyay, Debaditya Roy, Partha Pratim Das

Organizations: Indian Institute of Technology Kharagpur, West Bengal, INDIA · Ashoka University Sonipat, Haryana, INDIA

Abstract

Preserving intangible cultural dances rooted in centuries of tradition and governed by strict structural and symbolic rules presents unique challenges in the digital era. Among these, Bharatanatyam, a classical Indian dance form, stands out for its emphasis on codified adavus and precise key postures. Accurately generating these postures is crucial not only for maintaining anatomical and stylistic integrity, but also for enabling effective documentation, analysis, and transmission to broader global audiences through digital means. We propose a pose-aware generative framework integrated with a pose estimation module, guided by keypoint-based loss and pose consistency constraints. These supervisory signals ensure anatomical accuracy and stylistic integrity in the synthesized outputs. We evaluate four configurations: standard conditional generative adversarial network (cGAN), cGAN with pose supervision, conditional diffusion, and conditional diffusion with pose supervision. Each model is conditioned on key posture class labels and optimized to maintain geometric structure. In both cGAN and conditional diffusion settings, the integrated pose guidance aligns generated poses with ground-truth keypoint structures, promoting cultural fidelity. Our results demonstrate that incorporating pose supervision significantly enhances the quality, realism, and authenticity of generated Bharatanatyam postures. This framework provides a scalable approach for the digital preservation, education, and dissemination of traditional dance forms, enabling high-fidelity generation without compromising cultural precision. Code is available at https://github.com/jagidsh/Generating-Key-Postures-of-Bharatanatyam-Adavus-with-Pose-Estimation.

Figures & tables

Figure 1. Visual comparison of posture alignment across classical dance forms. a: Bharatanatyam postures illustrating vertical, horizontal, and diagonal axes and body symmetry (source ( Balaji, 2021 ) ). b: Kathak poses emphasizing upright torso and mirrored hand gestures (source ( Mallick, 2017 ) ). c: Ballet positions such as arabesque and plié, showing turnout and axial balance (source ( Daprati et al., 2009 ; Sarah, 2015 ) ). Despite cultural differences, these styles share a common reliance on geometrically structured key postures, justifying alignment-preserving approaches in generative modeling.

Figure 2. Illustration of Kuditta Tattal Adavu sequence in Bharatanatyam showcasing the structured progression of body postures. Each frame represents a temporally ordered key pose, reflecting the codified nature of movement transitions in the dance form.

Figure 3. Overview of the proposed model. A conditional generative model generates a dance image $I_{\text{gen}}$ from a latent code $z$ and a key posture label. Pose Estimation Module is used to extract 2D skeletal keypoints. Keypoint Loss – ( $\mathcal{L}_{\text{kp}}$ ) average squared Euclidean distance between normalized corresponding keypoints from $I_{\text{gen}}$ and $I_{\text{real}}$ , enforcing pixel-wise accuracy of each joint. Pose Consistency Loss – ( $\mathcal{L}_{\text{pose}}$ ) difference in internal pose structure capturing relative distances and angles between joints—between the generated and real poses.

Figure 4. Comparison of generated Adavu key postures across four models—(a) Conditional GAN (cGAN), (b) cGAN with Pose Estimation (CGPE), (c) Conditional Diffusion, and (d) Conditional Diffusion with Pose Estimation (CDMPE). Real reference images (ground truth) are shown alongside for comparison. While all models produce visually plausible results, the diffusion-based models demonstrate smoother textures and higher fidelity. The Conditional Diffusion with Pose Estimation (d) exhibits the most anatomically accurate and culturally faithful postures, preserving structural constraints and symbolic hand-gesture alignment essential to Bharatanatyam.

Figure 5. Complete set of generated Adavu key postures ( $\mathbf{128\times 128}$ ) using the Conditional Diffusion with Pose Estimation model. All 20 representative postures are synthesized to reflect structural precision, anatomical symmetry, and cultural authenticity specific to Bharatanatyam. This visualization highlights the model’s ability to generalize across a diverse set of codified movements while preserving stylistic consistency.

Model	FID ( $\downarrow$ )	MS-SSIM ( $\uparrow$ )
Conditional GAN (cGAN)	26.27	0.71
cGAN with Pose Estimation (CGPE)	24.28	0.73
Conditional Diffusion (cDiffusion)	21.88	0.75
cDiffusion with Pose Estimation (CDMPE)	19.32	0.78

Table 1. Impact of Pose Estimation Modules on Synthesis Quality

Model	FID ( $\downarrow$ )	MS-SSIM ( $\uparrow$ )
cGAN with $\mathcal{L}_{\text{kp}}$	24.77	0.72
cGAN with $\mathcal{L}_{\text{pose}}$	24.45	0.71
cDiffusion with $\mathcal{L}_{\text{kp}}$	20.15	0.76
cDiffusion with $\mathcal{L}_{\text{pose}}$	20.11	0.77
CDMPE (with $\mathcal{L}_{\text{kp}}$ and $\mathcal{L}_{\text{pose}}$ )	19.32	0.78

Table 2. Impact of Keypoint (

\mathcal{L}_{\text{kp}}

) and Pose consistency loss (

\mathcal{L}_{\text{pose}}

) on Synthesis Quality (ablation study)

Figure 6. FID score comparison for individual Bharatanatyam key postures across four generation models—cGAN, cGAN with pose estimation, cDiffusion, and cDiffusion with pose estimation. Lower FID(↓) values indicate better generation fidelity.

Figure 7. MS-SSIM scores for Bharatanatyam key postures across various generative models-—cGAN, cGAN with pose estimation, cDiffusion, and cDiffusion with pose estimation. Higher values (↑) represent better structural similarity and perceptual quality.

Method	FID ( $\downarrow$ )	MS-SSIM ( $\uparrow$ )
InsetGAN ( Frühstück et al., 2022 )	20.72	0.77
PCGAN ( Liang et al., 2019 )	21.17	0.75
XingGAN ( Tang et al., 2020 )	22.43	0.76
StyleGAN-Human ( Fu et al., 2022 )	20.32	0.74
PIDM ( Bhunia et al., 2022 )	23.57	0.71
CDMPE (Our Method)	19.32	0.78

Table 3. Quantitative comparison of key posture generation performance

Human Motion Generation

Human motion generation creates articulated movement from language, audio, video, spatial constraints, or interaction goals.

See all 291 papers

Sequence Recognition in Bharatnatyam dance
Sep 14, 2026Himadri Bhuyan, Rohit Dhaipule, Partha Pratim DasHuman Activity Recognition Video Action Recognition
Indian Institute of Technology Kharagpur, Kharagpur 721302, West Bengal, India
Mudragen: Geometrically Supervised Generation of Interacting Two-Hand Mudras for Preserving Indian Classical Dance Heritage
Sep 3, 2026Jagadish Kashinath Kamble, Jayanta Mukhopadhyay, Debaditya Roy +1Conditional Image Generation Human Motion Generation
Indian Institute of Technology Kharagpur Kharagpur, West Bengal, INDIA · Ashoka University Sonipat, Haryana, INDIA
DanceCrafter: Fine-Grained Text-Driven Controllable Dance Generation via Choreographic Syntax
Apr 20, 2026Hang Yuan, Xiaolin Hu, Yan Wan +8Text-to-Motion Generation Human Motion Generation
1East China Normal University · 5Zhongguancun Academy · 4New York University Shanghai +3

Sep 14, 2026J/K move · Enter open · S save

Sequence Recognition in Bharatnatyam dance

Himadri Bhuyan, Rohit Dhaipule, Partha Pratim Das

Indian Institute of Technology Kharagpur, Kharagpur 721302, West Bengal, India

Human Activity Recognition Video Action Recognition Sequence Alignment

Open paper AskPDF arXiv

Bharatanatyam is the oldest Indian Classical Dance (ICD) which is learned and practiced across India and the world. Adavu is the core of this dance form. There exist 15 Adavus and 58 variations. Each Adavu variation comprises a well-defined set of motions and postures (called dance steps) that occur in a particular order. So, while learning Adavus, students not only learn the dance steps but also take care of its sequence of occurrences. This paper proposed a method to recognize these sequences. In this work, firstly, we recognize the involved Key Postures (KPs) and motions in the Adavu using Convolutional Neural Network (CNN) and Support Vector Machine (SVM), respectively. In this, CNN achieves 99% and SVM's recognition accuracy becomes 84%. Next, we compare these KP and motion sequences with the ground truth to find the best match using the Edit Distance algorithm with an accuracy of 98%. The paper contributes hugely to the state-of-the-art in the form of digital heritage, dance tutoring system, and many more. The paper addresses three novelties; (a) Recognizing the sequences based on the KPs and motions rather than only KPs as reported in the earlier works. (b) The performance of the proposed work is measured by analyzing the prediction time per sequence. We also compare our proposed approach with the previous works that deal with the same problem statement. (c) It tests the scalability of the proposed approach by including all the Adavu variations, unlike the earlier literature, which uses only one/two variations.

Abstract

Human Motion Generation

Explore similar work

Sequence Recognition in Bharatnatyam dance

Mudragen: Geometrically Supervised Generation of Interacting Two-Hand Mudras for Preserving Indian Classical Dance Heritage

DanceCrafter: Fine-Grained Text-Driven Controllable Dance Generation via Choreographic Syntax