🚀 HiCrew

[HiCrew: Hierarchical Reasoning for Long-Form Video Understanding via Question-Aware Multi-Agent Collaboration]

🎉 Our work has been accepted by ICME 2026! 🎉

---

📖 Overview

The following image illustrates the case study and the overall method of our HiCrew approach:

Note: [Qualitative analysis on a Causal reasoning question from NExT-QA"]

---

⚙️ Installation

1. Clone the repository

git clone https://github.com/SYSUzzz/HiCrew.git
cd HiCrew

2. Set up the environment

conda create -n hicrew python=3.11 -y
conda activate hicrew

3. Install dependencies

pip install crewai
pip install openai                       
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
pip install pandas                      
pip install transformers==4.28.1         
pip install accelerate  
pip install opencv-python      
pip install numba               
pip install scikit-learn  

4. Update Kmeans-pytorch

git clone https://github.com/subhadarship/kmeans_pytorch
cd kmeans_pytorch

Please replace the init file in "kmeans_pytorch" folder with the file we provide in "./HybridTree/kmeans_pytorch" folder (this repo). And run the following command.

pip install --editable .

5. Download dataset

NExT-QA

EgoSchema

---

📊 Performance

Our approach has been extensively evaluated. Here are our experimental results compared with other models:

---

📝 Citation

If you find our work or this code useful for your research, please consider citing our paper:

@inproceedings{hicrew2026,
  title     = {[HiCrew: Hierarchical Reasoning for Long-Form Video Understanding via Question-Aware Multi-Agent Collaboration]},
  author    = {[Yuehan Zhu, Jingqi Zhao and Jiawen Zhao]},
  booktitle = {IEEE International Conference on Multimedia and Expo (ICME)},
  year      = {2026}
}

---

📄 License

This project is licensed under the MIT License