KidsGuard: Fine-Grained Approach for Child Unsafe Video Representation and Detection

This repository contains the code for the implementation of the paper titled KidsGuard: Fine-Grained Approach for Child Unsafe Video Representation and Detection by Singh et al. published at ACM SAC 2019.

Dataset

The dataset used for the paper can be from here: http://precog.iiitd.edu.in/requester.php?dataset=kidsguard2019

Experiment Steps

• Start by downloading the dataset.
• Download the YouTube videos using the video IDs mentioned in the dataset
• Once downloaded, use the notebooks in directory /extract_video to obtain video frames and then their VGG16 features.
• Use the notebooks in the /process_utils directory to parse annotations from the downloaded dataset, and aggregate clips and features for experiments.
• The notebooks in /train directory contain the notebooks to train the autoencoder and the classifier.
• /metrics contains the notebook to plot the training and testing results.

Dependencies

The project uses Python 3 dependencies explicitly, for processing and training. All the code is run on JupyterLab computational environment and Anaconda is used as a package manager as well as a virtual environment manager. All the dependencies are exported in the environment.yml file. Make a new environment using:

$ conda env create -f environment.yml

Citation

If you found this code or our paper useful, please consider citing the following paper:

@inproceedings{singh2019kidsguard,
    author = {
        Singh, Shubham and 
        Kaushal, Rishabh and 
        Buduru, Arun Balaji and 
        Kumaraguru, Ponnurangam
    },
    title = {{KidsGUARD: Fine Grained Approach for Child Unsafe Video Representation and Detection}},
    booktitle={Proceedings of the 34th Annual {ACM} Symposium on Applied Computing},
    location = {Limassol, Cyprus},
    year={2019}
}