Open Source AI Finder

A high-resolution dataset and a method for estimating 4D scene flow (3D motion of points over time) from multi-view camera inputs.

A dataset and baseline model for open-domain event extraction from conversations, aiming to identify and structure event information discussed in dialogues.

A technique that allows a language model to answer questions about a large document by ingesting its entire content into the context, using a leave-one-out attention mechanism, without needing to fine-tune the model.

A model that generates realistic and controllable human actions within 3D scenes based on natural language descriptions.

A physically-based rendering framework for synthesizing realistic hyperspectral images (HSI), which can be used as training data for other deep learning models.

A 1-trillion parameter Chinese-English bilingual large language model, demonstrating strong capabilities in both languages.

A training-free method to personalize text-to-image models, enabling the generation of images featuring specific subjects (like a person or pet) from just a few personal photos.

An AI model based on DeepVariant technology that accurately identifies genetic variants in tumors (somatic variants) from DNA sequencing data.

A unified diffusion model that generates high-fidelity 3D objects and consistent multi-view images from either a single image or a text prompt.

An efficient framework for Large Multimodal Models (LMMs) to process and understand long videos in a streaming fashion, maintaining high accuracy without needing to access the entire video at once.