Enhancing 3D Avatar Realism with Indian Clothing and Motion Capture Data

The Augmented Reality and Virtual Reality (AR/VR) industry has surged recently. Despite this, current work in 3D animation focuses on generating avatars or replicating motion on non-refined models lacking detail and personalization. Furthermore, 3D avatar personalization is limited to Western clothing, with Indian clothing largely unexplored. This proposal incorporates motion sequences into detailed 3D avatars adaptable to traditional Indian clothing. Initially, 3D avatars are generated from 2D images of people in Indian attire. Motion sequence data is extracted from monocular RGB videos and stored as Skinned Multi-Person Linear model (SMPL) parameters. The final stage integrates these components. For 3D model generation, Skinned Multi-Person Linear model - eXpressive (SMPL-X) is used, followed by Explicit Clothed humans Optimized via Normal integration (ECON). An iterative feedback loop with an optimizer and scheduler enhances the model. Motion sequence data uses a pre-trained CNN model, temporal encoder, and adversarial learning network. The Structural Similarity Index Metric (SSIM) evaluates the models. The final output creates personalized 3D animatable avatars with motion sequence data, benefiting the AR/VR and gaming industry.