Student Topics

AI-Based Detection of Explicit Content in Image and Video Data

Project/Bachelor thesis
Supervisor: Martin Kampel
Status: open

Motivation

Every day, enormous amounts of image and video content are uploaded to platforms like YouTube, Instagram, or X (formerly Twitter). Among this content — depending on the platform and moderation policies — are also materials containing nudity, pornography, or other explicit material. Such content not only violates the guidelines of many online platforms but also contravenes legal regulations, particularly referring to §§ 94 paras. 3 and 4, § 95, and § 96a of the Austrian Criminal Code (StVG).
As part of the STeRn project, an AI-based system for the detection of explicit content is being developed that considers both legal and technical requirements. The goal is to create a robust, modularly integrable algorithm that can be embedded into community-based platforms.

Objective of the Project

The aim of the project is the design, development, and evaluation of a real-time system for detecting explicit image and video content. The approach involves combining Convolutional Neural Networks (CNNs) for analyzing visual features with Long Short-Term Memory (LSTM) networks for processing temporal context to enable reliable decisions about the presence of undesirable content.

Particular focus will be on:

  • Robust detection under difficult lighting conditions, perspectives, or partial occlusions

  • Minimizing false positives/negatives to preserve legitimate content

  • Potential integration into an existing platform for community use

Tasks

  • Literature review on methods for explicit content detection (CNNs, RNNs, LSTMs)

  • Selection of appropriate, ethically permissible datasets for training and validation (e.g., NPDI, Pornography-800, Safe/Unsafe Benchmarks)

  • Development of a multimodal detection model (e.g., CNN + LSTM architecture)

  • Training and evaluation of the model with respect to precision, recall, and F1-score

  • (Optional) Implementation of real-time inference on videos (e.g., using OpenCV or FFmpeg)

  • (Optional) Integration into a sample platform with GUI feedback

Required Skills

  • Good knowledge of Python and deep learning

  • Experience with PyTorch or TensorFlow

  • Fundamentals in computer vision and neural networks

  • Interest in ethical, legal, and societal issues related to AI

Contact: martin.kampel@tuwien.ac.at