Concatenation of Attention Enhanced Spatial and Temporal Features for Violence Detection from Videos

This work is an attempt towards simulating how humans react to violent situations using Convolutional Neural Network (CNN) and Long Short Term Memory (LSTM) architectures.

Dataset: Hockey Fights

Dataset Directory Structure:

───HockeyFight
    ├───train
    │   ├───Fight
    │   │       Fight_1.avi
    │   │       Fight_2.avi
    │   │
    │   └───No_Fight
    │           No_Fight_1.avi
    │           No_Fight_2.avi
    └───val
        ├───Fight
        │       Fight_1.avi
        │       Fight_1.avi
        │
        └───No_Fight
                No_Fight_1.avi
                No_Fight_2.avi

---

Try it yourself (Windows):

1. Clone the repository:

    https://github.com/004Ajay/Main-Project.git

2. Navigate to the project directory:

    cd Main-Project

3. Create a virtual environment or see this for a detailed guide.

    python -m venv <your env name>

    cd <your env name>

    Scripts\activate

4. Install the required dependencies

    pip install -r requirements.txt

dependecies does not include libraries for the files in the Experiments Folder.

5. Download the dataset and arrange as per the structure given above.

6. Start training:

    python train.py

7. Evaluate your model:

    python evaluate.py

---

Documents

• Main Project Report Phase 1

• Main Project Report Phase 2

---

Poster

---

Creators

Ajay T Shaju | Emil Saj Abraham | Justin Thomas Jo | Vishnuprasad KG