OAIR

Opus AI Research

Who are we?

Opus AI Research is the research division of OpusClip, dedicated to advancing artificial intelligence technologies, particularly in video understanding. Our team has contributed numerous high-impact publications in top-tier conferences and journals, covering areas such as video editing, generative models, and few-shot learning. Additionally, we actively participate in international AI competitions, achieving top rankings in video understanding challenges.

Research Gallery

AI Research Visualization
RSVP: Reasoning Segmentation via Visual Prompting and Multi-modal Chain-of-Thought
AI Research Visualization
VEU-Bench: Towards Comprehensive Understanding of Video Editing
Video Understanding Research
Number it: Temporal Grounding Videos like Flipping Manga
Generative AI Research
Video Repurposing from User Generated Content: A Large-scale Dataset and Benchmark
Generative AI Research
Frame Order Matters: A Temporal Sequence-Aware Model for Few-Shot Action Recognition

Team Achievements

article

Publications

Title Conference
RSVP: Reasoning Segmentation via Visual Prompting and Multi-modal Chain-of-Thought ACL 2025 Main
VEU-Bench: Towards Comprehensive Understanding of Video Editing CVPR 2025 Highlight
Number it: Temporal Grounding Videos like Flipping Manga CVPR 2025 Poster
Video Repurposing from User Generated Content: A Large-scale Dataset and Benchmark AAAI 2025 Poster
Unlearning Concepts in Diffusion Model via Concept Domain Correction and Concept Preserving Gradient AAAI 2025 Poster
Envisioning Class Entity Reasoning by Large Language Models for Few-shot Learning AAAI 2025 Poster
Frame Order Matters: A Temporal Sequence-Aware Model for Few-Shot Action Recognition AAAI 2025 Poster
Object-Level Out-of-Distribution Detection via Regional Uncertainty Alignment of Multimodal Representations AAAI 2025 Oral
VisTa: Visual-contextual and Text-augmented Zero-shot Object-level OOD Detection ICASSP 2025 Poster
Cocktail: Chunk-Adaptive Mixed-Precision Quantization for Long-Context LLM Inference DATE 2025 Best Paper
Exploring Diverse In-Context Configurations for Image Caption NeurIPS 2023 Poster
Few-Shot Font Generation by Learning Fine-Grained Local Styles CVPR 2022 Poster
Exploiting Deep Cross-Slice Features From CT Images For Multi-Class Pneumonia Classification ICIP 2021 Best Student Paper
Fuzzy SLIC: Fuzzy Simple Linear Iterative Clustering TCSVT 2020

Competitions & Awards

Winner

VidLLMs Workshop@CVPR 2025

Multilingual Video Reasoning Evaluation Challenge

Runner-up

VidLLMs Workshop@CVPR 2025

Complex Video Reasoning & Robustness Evaluation Challenge

Third Place

SoccerNet Challenges@CVPR 2025

Multi-View Foul Recognition Track

Winner

LOVEU Workshop@CVPR 2024

Long-Term Video QA Track

Winner

Perception Challenge@ECCV 2024

Hour-Long Video QA Track

Runner-up

KDD Cup 2020

E-Commerce Debiasing Challenge