Computational Imaging and Visual Innovations (CIVI)

The Center for Computational Imaging and Visual Innovations (CIVI) specializes in research projects centered towards solving real-world problems through deep learning solutions trained on visual data. Research projects under CIVI focus on specific goals, particularly (1) data collection, (2) model building and optimization, and (3) application development.

Research Team

  • Lab Head: Arren Matthew Antioquia (M.Sc. Computer Science, National Taiwan University of Science and Technology, De La Salle University)
  • Gabriel Avelino Sampedro (M.Sc. Computer Engineering, Mapua University)
  • Richmond Manga (M.Sc. Computer Science, National Taiwan University of Science and Technology, De La Salle University)

International Collaborators

  • Daniel Stanley Tan (Ph.D. Computer Science, National Taiwan University of Science and Technology)
  • John Jethro Virtusio (Ph.D. Computer Science, National Taiwan University of Science and Technology)

Specialization

Data Collection

Proper collection and labeling of image datasets are active areas of research in computer vision. Current research papers propose larger and better datasets to improve the performance of deep neural networks on specific visual tasks.

This is an important research area under CIVI, especially since deep learning solutions are heavily reliant on high-quality and large-scale datasets. More specifically, research projects under this program are tasked to:

  1. Identify the limitations of existing datasets and/or data collection methods on a specific domain.
  2. Introduce improvements by proposing a novel high-quality dataset that addresses the issues of current datasets.
  3. Evaluate latest models on the proposed dataset.

 

Model Building and Optimization

Models are often designed to improve their performance by incorporating more computations. However, this method usually results in larger models with slower inference time and larger space requirements, which are not always applicable and practical for all use cases.

Projects under CIVI are designed to develop models based on their intended applications and constraints. Moreover, proposed models should improve the performance of the latest models on certain metrics. Research projects under this program are tasked to:

  1. Identify a specific target domain.
  2. Analyze the limitations of the latest models designed for the target domain.
  3. Develop a model to address the limitations of current methods.
  4. Compare the performance of the proposed model to state-of-the-art methods designed for the target domain.

 

Application Development


Most trained models are not utilized since they are not integrated into useful apps. Thus, developing applications is necessary in utilizing deep learning solutions to real-world problems.

Applications serve as the interface of the trained model to its target users. Research projects under this program are tasked to:

  1. Identify models that satisfy resource constraints in target platform and use case.
  2. Develop an app that will aid in utilizing selected models to solve practical issues.
  3. Assess the app using various applicable performance metrics.