Evaluating Unity-Generated Data for Real-World Human Detection

Evaluating Unity-Generated Data for Real-World Human Detection

Supervisor: Dr Zhuoling Huang 

What you will be doing 

This internship project will focus on evaluating the effectiveness of human data generated using Unity Asset Store models for neural network training in computer vision tasks. This project aims to leverage pre-made human models to create synthetic datasets and compare their performance against traditional real human datasets.  

Figure 1: examples of free human models in Unity (https://assetstore.unity.com/packages/3d/characters/humanoids/suit-character-pack-generic-16772 and https://assetstore.unity.com/packages/3d/characters/humanoids/sci-fi/sci-fi-post-apocalyptic-young-male-character-221414)

You will be involved in selecting suitable human models from the Unity Asset Store, understanding their functionalities, and using Unity’s animation capabilities to generate relevant data for tasks like pose estimation or facial recognition. Your work will also involve using existing neural network architectures, such as YOLO, to train on the synthetic data. You will preprocess this data, conduct training, and compare it with real human datasets from publicly available sources.  

Throughout the project, you will gain hands-on experience in machine learning, neural network training, and data preprocessing. You will also develop skills in using Unity for data generation, animation, and scripting. This project will provide insights into the accuracy, generalizability, and limitations of using synthetic data for computer vision applications, contributing to the advancement of user-friendly methods in this field.   

What skills would be useful to have for this project 

  • Programming skills in Python
  • Machine learning tools related to computer vision (e.g. Keras, TensorFlow, PyTorch)
  • C# would be helpful, but it is not required.   

How to get more information and apply 

The prospective candidates can apply by sending an email to Zhuoling (zhuang@lincoln.ac.uk) and providing a short background summary focusing on relevant interests and skills together with a CV. All candidates meeting the skills requirements will be accept and in case of multiple expressions of interest, the project scope will be negotiated with individual candidates. The application deadline is 30th of June and the candidates will be informed about the selection outcome by the 4th of July with an anticipated project start on the 8th of July (this is negotiable).