HumanML3D
PulseAugur coverage of HumanML3D — every cluster mentioning HumanML3D across labs, papers, and developer communities, ranked by signal.
3 day(s) with sentiment data
-
New IKMo framework generates motion from images with diverse constraints
Researchers have introduced IKMo, a novel framework for generating motion from images, addressing the challenge of integrating diverse spatial constraints. The system utilizes TPMDM, a diffusion model with a parallel co…
-
LLMs advance motion understanding with geometric alignment and timestamped poses · 2 sources tracked
Researchers have developed new frameworks for large-language models (LLMs) to better understand human motion. GeoMotionGPT focuses on aligning the geometry of motion with the LLM's embedding space through orthogonality,…
-
Research paper questions shared latent space assumptions in motion VAEs
A new research paper explores the concept of shared latent spaces in motion VAEs, challenging the assumption that shared parameters or dimensionality automatically lead to directly interchangeable representations. The s…
-
T2M Mamba enhances text-driven 3D motion generation by coupling periodicity and saliency
Researchers have introduced T2M Mamba, a novel approach to text-driven 3D human motion generation. This method addresses limitations in existing models by coupling motion periodicity and keyframe saliency, rather than t…
-
New HiSTF Mamba framework enhances text-driven motion generation
Researchers have introduced HiSTF Mamba, a novel framework for text-driven motion generation that addresses limitations in existing methods. The proposed system utilizes a linear-time approach with parallel part-based a…
-
BiPO network advances text-to-motion synthesis with bidirectional approach
Researchers have developed BiPO, a Bidirectional Partial Occlusion Network designed to improve text-to-motion synthesis. This novel model integrates part-based generation with a bidirectional autoregressive architecture…
-
New TACD method trains efficient text-to-motion models
Researchers have developed a new method called Terminal-Amplification-Controlled Distillation (TACD) to create more efficient text-to-motion models. This on-policy approach trains smaller motion generators using text pr…
-
FloodDiffusion 2 advances streaming motion generation with efficiency and path control
Researchers have introduced FloodDiffusion 2 (FD2), an advanced framework designed for efficient and controllable streaming motion generation. This new model builds upon its predecessor, FloodDiffusion (FD1), by address…
-
New Triangular Resampling method improves long-horizon motion generation
Researchers have developed Triangular Resampling (TR), a novel post-training technique designed to improve the accuracy of motion diffusion models over long sequences. This method addresses the issue of error accumulati…
-
BiMoGen framework uses diffusion for bidirectional motion-text generation
Researchers have developed BiMoGen, a novel framework for bidirectional motion-text generation that utilizes masked discrete diffusion. This approach addresses limitations of previous autoregressive models by enabling i…
-
UniMo framework unifies human and animal 3D motion generation
Researchers have developed UniMo, a novel framework for generating 3D motion that unifies human and animal movements. The system addresses the challenge of diverse animal skeletal structures by converting parametric ske…
-
ReMoMask-2 advances text-to-motion generation with structure-aware retrieval
Researchers have introduced ReMoMask-2, a novel framework designed to enhance text-to-motion generation by addressing challenges in retrieval and fusion mechanisms. The system employs Hierarchical Bidirectional Momentum…
-
ReMoMask-2 advances text-to-motion generation with latent retrieval
Researchers have introduced ReMoMask-2, an advancement in text-to-motion generation that directly embeds retrieval into the generator's latent space. This approach addresses the representation gap found in previous mode…
-
New DeMoDiff model enhances human motion generation with part-level control
Researchers have developed a new framework called DeMoDiff for generating human motion. This model addresses limitations in existing methods by using a spatiotemporal variational auto-encoder to encode individual body j…
-
Rectified Flow Outperforms Diffusion in MotionGPT3 for Text-to-Motion Generation
Researchers have conducted a study comparing diffusion and rectified flow objectives within the MotionGPT3 framework for text-driven motion generation. The experiments, performed on the HumanML3D dataset, indicate that …
-
WaMo framework enhances text-motion retrieval with wavelet analysis · arXiv cs.CV
Researchers have developed WaMo, a new framework for text-motion retrieval that uses wavelet decomposition to analyze 3D motion sequences. This method captures joint-specific and time-varying motion details at multiple …
-
MUGEN framework unifies motion understanding and generation with continuous latents
Researchers have introduced MUGEN, a novel unified framework designed for efficient motion understanding and generation. Unlike previous methods that rely on discrete motion codebooks, MUGEN utilizes a single draw from …
-
New framework KineMIC enhances few-shot action synthesis for HAR
Researchers have developed KineMIC, a novel transfer learning framework designed to improve few-shot action synthesis for skeletal-based Human Activity Recognition (HAR). This method adapts text-to-motion diffusion mode…
-
ScaleMoGen framework advances text-driven human motion generation
Researchers have introduced ScaleMoGen, a novel framework for generating human motion from text descriptions. This approach utilizes an autoregressive method that predicts motion tokens across multiple scales, from coar…
-
New research refines diffusion model noise for better video generation control
Two new research papers propose novel methods for improving controllability in diffusion-based video generation by manipulating the initial noise input. The first paper, WINRO, focuses on text-to-motion generation by re…