Large Motion Model for Unified Multi-Modal Motion Generation

The Large Motion Model (LMM) unifies various motion generation tasks into a scalable, generalist model, demonstrating broad applicability and strong generalization across diverse tasks.

Digital Life Project: Autonomous 3D Characters with Social Intelligence

Digital Life Project is a framework utilizing language as the universal medium to build autonomous 3D characters, who are capable of engaging in social interactions and expressing with articulated body motions, thereby simulating life in a digital environment.

FineMoGen: Fine-Grained Spatio-Temporal Motion Generation and Editing

FineMoGen is a diffusion-based motion generation and editing framework that can synthesize fine-grained motions, with spatial-temporal composition to the user instructions.

InsActor: Instruction-driven Physics-based Characters

InsActor is a principled generative framework that leverages recent advancements in diffusion-based human motion models to produce instruction-driven animations of physics-based characters.

SMPLer-X: Scaling Up Expressive Human Pose and Shape Estimation

SMPLer-X is the first generalist foundation model for Expressive human pose and shape estimation (EHPS). With big data and the large model, SMPLer-X exhibits strong performance across diverse test benchmarks and excellent transferability to even unseen environments.

PointHPS: Cascaded 3D Human Pose and Shape Estimation from Point Clouds

PointHPS iteratively refines point features through a cascaded architecture to achieve more accurate 3D Human pose and shape estimation(HPS) from point clouds.

ReMoDiffuse: Retrieval-Augmented Motion Diffusion Model

ReMoDiffuse is a retrieval-augmented 3D human motion diffusion model. Benefiting from the extra knowledge from the retrieved samples, ReMoDiffuse is able to achieve high-fidelity on the given prompts.

BiBench: Benchmarking and Analyzing Network Binarization

A rigorously designed benchmark with in-depth analysis for network binarization. It first carefully scrutinizes the requirements of binarization in the actual production and define evaluation tracks and metrics for a comprehensive and fair investigation.

MotionDiffuse: Text-Driven Human Motion Generation with Diffusion Model

The first text-driven motion generation pipeline based on diffusion models with probabilistic mapping, realistic synthesis and multi-level manipulation ability.

HuMMan: Multi-Modal 4D Human Dataset for Versatile Sensing and Modeling

A large-scale multi-modal(color images, point clouds, keypoints, SMPL parameters, and textured meshes) 4D human dataset with 1000 human subjects, 400k sequences and 60M frames.

AvatarCLIP: Zero-Shot Text-Driven Generation and Animation of 3D Avatars

AvatarCLIP is the first zero-shot text-driven pipeline, which empowers layman users to generate and animate 3D Avatars by natural language description.

Balanced MSE for Imbalanced Visual Regression

A statistically principled loss function to address the train/test mismatch in imbalanced regression, coincides with the supervised contrastive loss.

Delving Deep into the Generalization of Vision Transformers under Distribution Shifts

A systematical comparison of the generalization ability between CNNs and ViTs. Three representative generalization-enhancement techniques are applied to ViTs to further explore their inner properties.

Playing for 3D Human Recovery

A large-scale synthetic human dataset collected using GTA-5 game engine, providing stable performance boost to both frame-based and video-based HMR.

BiBERT: Accurate Fully Binarized BERT

BiBERT is the first fully binarized BERT. It introduces an efficient Bi-Attention structure and a DMD scheme, which yields impressive 59.2x and 31.2x saving on FLOPs and model size.

REFINE: Prediction Fusion Network for Panoptic Segmentation

REFINE achieves high-quality panoptic segmentation by improving cross-task prediction fusion, and within-task prediction fusion.

CSG-Stump: A Learning Friendly CSG-Like Representation for Interpretable Shape Parsing

CSG-Stump learns shapes from point clouds and discovers the underlying constituent modeling primitives and operations.

Towards Overcoming False Positives in Visual Relationship Detection

SABRA explores the imbalanced distribution in Human-Object Interaction detection. It further proposes a new pipeline to equip the model with sufficient spatial information.

BiPointNet: Binary Neural Network for Point Clouds

BiPointNet is the first fully binarized network for point cloud learning. BiPointNet gives an impressive 14.7x speedup and 18.9x storage saving on real-world resource-constrained devices.

Efficient Attention: Attention with Linear Complexities

Efficient Attention reduces the memory and computational complexities of the attention mechanism from quadratic to linear. It demonstrates significant improvement in performance-cost trade-offs on a variety of tasks including object detection, instance segmentation, stereo depth estimation, and temporal action lcoalization.

Graph attention based proposal 3d convnets for action detection

AGCN-P-3DCNNs fuses intra and inter attention to model intra long-range dependencies and inter dependencies simultaneously. It also contains a simple and effective framewise classifier, which enhances the feature presentation capabilities of backbone model.