Optimizing SAGE Net: Sequential Training of Stratified Diffusion Models and Full-Body Decoder
22 Oct 2025
This article provides the implementation specifics for the Stratified Diffusion models in SAGE Net.
SAGE Net Ablation Study: Analyzing the Impact of Input Sequence Length on Performance
22 Oct 2025
This article presents an ablation study on the SAGE Net design, focusing on the critical factor of input sequence length (N) for online inference.
Disentangling Upper and Lower Body Motion: The Key Finding for Full-Body Avatar Reconstruction
22 Oct 2025
This conclusion summarizes the key finding that disentangling upper and lower body motions is crucial for accurate 3D avatar generation
The Importance of Disentanglement: SAGE Outperforms Unified VQ-VAE Baselines in Full-Body Motion
22 Oct 2025
This article presents an ablation study confirming that disentangling motion latents into upper and lower halves significantly enhances 3D avatar reconstruction
Quantitative and Qualitative Results: SAGE Outperforms SOTA in Full-Body 3D Avatar Reconstruction
22 Oct 2025
This article presents the quantitative and qualitative results for the SAGE model across three evaluation settings
Quantifying Motion Accuracy: MPJRE, MPJPE, and Consistency Metrics for HMD-based Avatar Generation
22 Oct 2025
This article outlines the evaluation metrics used to benchmark 3D avatar reconstruction models on the AMASS motion capture dataset.
Temporal Refinement in Stratified Motion Diffusion: Utilizing GRU for Smoothed Full-Body Prediction
22 Oct 2025
This article describes the online inference stage of the model, which predicts 3D avatars in a frame-by-frame manner using a sliding window approach
Generating Consistent Full-Body Avatars: Stratified Motion Diffusion for Decoupled Kinematics
22 Oct 2025
This article introduces Stratified Motion Diffusion, a novel technique that employs cascaded latent sampling to reconstruct full-body avatars.
Disentangled Motion Representation: Encoding Full-Body Avatars into Discrete Latent Spaces
22 Oct 2025
This article describes the technique for achieving disentangled motion representation by splitting full-body kinematics into upper and lower halves.