Ego-4D

Ego-4D is a large-scale wearable-egocentric dataset designed to evaluate proactive procedural assistance systems, featuring explicit Out-of-Plan annotations and recovery steps.

Papers using Ego-4D (10)

Plan, Watch, Recover: A Benchmark and Architectures for Proactive Procedural Assistance2026Recurrent Reasoning with Vision-Language Models for Estimating Long-Horizon Embodied Task Progress2026EgoSound: Benchmarking Sound Understanding in Egocentric Videos2026EVA02-AT: Egocentric Video-Language Understanding with Spatial-Temporal Rotary Positional Embeddings and Symmetric Optimization2025Know-show: Benchmarking Video-language Models On Spatio-temporal Grounded Reasoning2025Vision and Intention Boost Large Language Model in Long-Term Action Anticipation2025EmbodiedGPT: Vision-Language Pre-Training via Embodied Chain of Thought2023 · 41 citesHourVideo: 1-Hour Video-Language Understanding2024 · 5 citesVLM Agents Generate Their Own Memories: Distilling Experience into Embodied Programs of Thought2024 · 2 citesX-MIC: Cross-Modal Instance Conditioning for Egocentric Action Generalization2024