🚀🚀 Efficient implementations of Native Sparse Attention

Official repo for paper "SceneGen: Single-Image 3D Scene Generation in One Feedforward Pass"

Development repository for the Triton language and compiler

SLA: Beyond Sparsity in Diffusion Transformers via Fine-Tunable Sparse–Linear Attention

[NeurIPS 2025] PhysCtrl: Generative Physics for Controllable and Physics-Grounded Video Generation

DiffusionNFT: Online Diffusion Reinforcement with Forward Process

Official repo of paper "Reconstruction Alignment Improves Unified Multimodal Models". Unlocking the Massive Zero-shot Potential in Unified Multimodal Models through Self-supervised Learning.

MapAnything: Universal Feed-Forward Metric 3D Reconstruction

SpatialVID: A Large-Scale Video Dataset with Spatial Annotations

[PG2025] PaMO: Parallel Mesh Optimization for Intersection-Free Low-Poly Modeling on the GPU

A video foundation model for unified Text-to-Video (T2V) and Image-to-Video (I2V) generation.

Import a 3D Model and automatically assign and export animations

A Survey of Efficient Attention Methods: Hardware-efficient, Sparse, Compact, and Linear Attention

A Python framework for accelerated simulation, data generation and spatial computing.

[NeurIPS 2025 Spotlight] Official repository for “Puppeteer: Rig and Animate Your 3D Models”

Reference PyTorch implementation and models for DINOv3

ViPE: Video Pose Engine for Geometric 3D Perception

Efficient triton implementation of Native Sparse Attention.

Qwen-Image is a powerful image generation foundation model capable of complex text rendering and precise image editing.

gpt-oss-120b and gpt-oss-20b are two open-weight language models by OpenAI

🔥 Official impl. of "DetailFlow: 1D Coarse-to-Fine Autoregressive Image Generation via Next-Detail Prediction"

Wan: Open and Advanced Large-Scale Video Generative Models

Lumos Project: Frontier generative model research by Alibaba DAMO Academy, including Lumos-1, etc.

Code for Streaming 4D Visual Geometry Transformer

Code of π^3: Permutation-Equivariant Visual Geometry Learning

Cameras as Relative Positional Encoding

CoPart (ICCV 2025): A part-based 3D generation framework & the first large-scale part-level 3D dataset.