Computer Vision

Scaling video generation models doesn’t guarantee they’ll learn physics; this study reveals they prioritize visual cues over true physical understanding.

GenXD: A unified model generating high-quality 3D & 4D scenes from any number of images, advancing the field of dynamic scene generation.

Adaptive Caching (AdaCache) dramatically speeds up video generation with diffusion transformers by cleverly caching and reusing computations, tailoring the process to each video’s complexity and motio…

DreamPolish: A new text-to-3D model generates highly detailed 3D objects with polished surfaces and realistic textures using progressive geometry refinement and a novel domain score distillation tech…

Randomized Autoregressive Modeling (RAR) sets a new state-of-the-art in image generation by cleverly introducing randomness during training to improve the model’s ability to learn from bidirectional c…

Constant Acceleration Flow (CAF) dramatically speeds up diffusion model generation by using a constant acceleration equation, outperforming state-of-the-art methods with improved accuracy and few-step…

High-performing video representation models can be trained using only synthetic videos and images, eliminating the need for large natural video datasets.

In-Context LoRA empowers existing text-to-image models for high-fidelity multi-image generation by simply concatenating images and using minimal task-specific LoRA tuning.

DELTA: A new method efficiently tracks every pixel in 3D space from monocular videos, enabling accurate motion estimation across entire videos with state-of-the-art accuracy and over 8x speed improvem…

HelloMeme enhances text-to-image models by integrating spatial knitting attentions, enabling high-fidelity meme video generation while preserving model generalization.

DynaMath, a novel benchmark, reveals that state-of-the-art VLMs struggle with variations of simple math problems, showcasing their reasoning fragility. It offers 501 high-quality seed questions, dyna…