--- language: - en library_name: transformers tags: - ego4d - egocentric-video - long-term-action-anticipation - intention-reasoning - grpo --- # INSIGHT (AAAI 2026) This repository provides the **Stage 2 model weights** for **INSIGHT: Intention-Guided Cognitive Reasoning for Egocentric Long-Term Action Anticipation** (AAAI 2026). The model is based on **Qwen2.5-VL-7B**, trained with **GRPO on Ego4D**. - **Code:** [iLearn-Lab/AAAI26-INSIGHT](https://github.com/iLearn-Lab/AAAI26-INSIGHT) - **Paper:** [INSIGHT](https://arxiv.org/abs/2508.01742)