Computer Vision

MagicMotion: A controllable video generation framework enabling precise object motion control through dense-to-sparse trajectory guidance.

InfU: A new framework for flexible photo re-creation while preserving identity using Diffusion Transformers(DiTs).

Coarse-to-Fine Token Prediction improves autoregressive image generation by assigning the same coarse label for similar tokens, balancing generation quality and computational efficiency.

GFS-VL: Enhancing few-shot 3D segmentation by synergizing vision-language models with few-shot learning for robust real-world application.

Expert Race: A flexible routing strategy for scaling diffusion transformer with mixture of experts.

CLS-RL: Rule-based RL tackles catastrophic forgetting in MLLM image classification, outperforming SFT with better generalization and efficiency.

TokenBridge bridges continuous and discrete tokens for autoregressive visual generation, achieving high-quality synthesis with simple autoregressive modeling.

4DGS-1K: Achieves 1000+ FPS for dynamic scene rendering via a compact, memory-efficient framework, offering a 41x storage reduction and 9x faster speed.

FluxFlow: Make your video generator stronger via temporal regularization!

FakeVLM: A multimodal model & artifact-annotated dataset for detecting synthetic images with interpretable explanations, setting a new benchmark.

MotionStreamer: Streaming motion generation w/ diffusion-based autoregressive model in causal latent space.

LEGION: Grounding and explaining synthetic image detection and refinement via multimodal learning.

Personalize diffusion models efficiently on devices without backpropagation.

DeepMesh: RL-guided auto-regressive creation of artist-quality 3D meshes, enhanced by tokenization & DPO for human-aligned aesthetics.

Roblox presents Cube, a 3D intelligence model using shape tokenization for text-to-shape, shape-to-text, and text-to-scene generation.

FluxViT: Flexible video models via adaptive token selection for efficient deployment!

MagicComp: Dual-Phase Refinement Enables Training-Free Compositional Video Generation

Impossible videos expose AI limits!

DiffMoE: Dynamically selects tokens for scalable diffusion transformers, unlocking new efficiency levels in image generation.

Cosmos-Transfer1: An adaptable conditional world generation model using multimodal control.