🏃‍♂️ Group Activity Recognition (Collective Activity Dataset)

📄 Project Overview

This project implements a Hierarchical CNN + LSTM pipeline for recognizing group activities from video sequences, inspired by the paper:

"A Hierarchical Deep Temporal Model for Group Activity Recognition"
Mostafa S. Ibrahim, Shengyu Zhang, Stan Sclaroff, Margrit Betke
https://arxiv.org/abs/1511.06040

Dataset:

Collective Activity Dataset (University of Michigan / Stanford CVGL)
Activities: Crossing, Waiting, Queueing, Walking, Talking

---

🚀 Pipeline Architecture (CNN + LSTM)

---

🔧 Installation

pip install -r requirements.txt

---

🛠️ Training

1️⃣ Person-Level Action Recognition

python scripts/train_person_model.py

2️⃣ Group-Level Activity Recognition

python scripts/train_group_model.py

---

📊 Evaluation

python scripts/evaluate_group_model.py