Simple 3D CNN for Action Recognition

Setup

Prerequisites

• Python 3.11
• NVIDIA GPU with CUDA support
• Miniconda or Anaconda

Environment Setup

1. Create and activate a new conda environment:

conda create -n learning python=3.11
conda activate learning

2. Install PyTorch and related packages:

conda install pytorch torchvision torchaudio pytorch-cuda=12.1 -c pytorch -c nvidia

3. Install other dependencies:

pip install -r requirements.txt

Project Structure

• dataset.py: Implements the SimpleVideoDataset class for video data handling
• model.py: Contains the Simple3DCNN architecture
• training.py: Training loop implementation
• test_model.py: Model evaluation script
• visualize.py: Visualization script to measure model performance and generate confusion matrix
• utils.py: Helper functions
• outputs/: Directory for storing model checkpoints and logs

Usage

Training

Hyperparameters

Update the training.py script to take in the correct hyperparameters

• batch_size: number of videos per batch
• learning_rate: learning rate for the optimizer
• num_epochs: number of epochs to train for
• patience: number of epochs to wait before early stopping (0 to disable)

Also update the dataset root directory in the training.py script

• dataset_root: root directory of the dataset

Dataset files:

• train.csv: training dataset
• val.csv: validation dataset
• test.csv: test dataset

Format of the dataset files(csv):

<video_path>, <label>
<video_path>, <label>
<video_path>, <label>
...

Training Script

python training.py

License

[Add your license information here]

Contact

[Add your contact information if you want]

#TODO update script to take true csvs instead of weird txt format