readme.md

Violence Detection using Conv3D

Model Architecture

• Type: 3D Convolutional Neural Network (Conv3D)
• Input: Video sequence of 16 frames, resized to 112x112.
• Structure:
  • 4 Conv3D Layers with BatchNorm, ReLU, and MaxPooling.
  • Flatten Layer.
  • 2 Fully Connected Layers.
  • Dropout (0.5) for regularization.
• Output: Binary Classification (Violence vs No-Violence).

Dataset Structure

The code expects a Dataset folder in the parent directory (or modify DATASET_DIR in train.py). Structure:

Dataset/
├── violence/
│   ├── video1.mp4
│   └── ...
└── no-violence/
    ├── video2.mp4
    └── ...

How to Run

1. Install dependencies: torch, opencv-python, scikit-learn, numpy.
2. Run python train.py.