long-video-understanding

[KDD'2026] "VideoRAG: Chat with Your Videos"

NVIDIA AI Blueprint for video search and summarization (VSS) is a GPU-accelerated reference architecture for building video analytics agents with real-time verified alerts, visual Q&A, and automated reporting. The VSS Blueprint uses vision language models (VLMs) such as NVIDIA Cosmos, LLMs such as NVIDIA Nemotron, RAG, and NVIDIA NIMs.

An open-weight 11B model series for long-form and real-time video understanding

[CVPR 2024] MovieChat: From Dense Token to Sparse Memory for Long Video Understanding

Official code for Goldfish model for long video understanding and MiniGPT4-video for short video understanding

✨✨[NeurIPS 2025] This is the official implementation of our paper "Video-RAG: Visually-aligned Retrieval-Augmented Long Video Comprehension"

🔥🔥MLVU: Multi-task Long Video Understanding Benchmark

[CVPR 2026] LongVT: Incentivizing "Thinking with Long Videos" via Native Tool Calling

Multi-granularity Correspondence Learning from Long-term Noisy Videos [ICLR 2024, Oral]

Context-Aware RAG library for Knowledge Graph ingestion and retrieval functions.

OmniAgent (ICML 2026): the first native omni-modal agent for active video perception — a 7B agent that beats Qwen2.5-VL-72B with 73% fewer frames on LVBench.

[ICLR 2025] TimeSuite: Improving MLLMs for Long Video Understanding via Grounded Tuning

LongVALE: Vision-Audio-Language-Event Benchmark Towards Time-Aware Omni-Modal Perception of Long Videos. (CVPR 2025))

ParaVT: Taming the Tool Prior Paradox for Parallel Tool Use in Agentic Video Reinforcement Learning

[EMNLP 2023] TESTA: Temporal-Spatial Token Aggregation for Long-form Video-Language Understanding

This is the official implementation of ReVisionLLM: Recursive Vision-Language Model for Temporal Grounding in Hour-Long Videos

Light-Omni: Reflex over Reasoning in Agentic Video Understanding with Long-Term Memory

Code for our ACL 2025 paper "Language Repository for Long Video Understanding"

[CVPR 2026] Wavelet-based Frame Selection by Detecting Semantic Boundary for Long Video Understanding

[NeurIPS 2025] HoPE: Hybrid of Position Embedding for Long Context Vision-Language Models