Research program: image description methods

The goal of the Image Description Methods research program is to develop advanced methods for generating logical, coherent situational descriptions from visual data. The program focuses on integrating vision systems, semantic segmentation, artificial intelligence methods and logic systems to create comprehensive tools for analyzing and interpreting images.

research area

Vision systems and semantic segmentation:

• Development of semantic and panoptic segmentation algorithms to separate objects and spatial relationships in visual scenes.
• Study of the effect of segmentation quality on the generation of situational descriptions.

AI methods and logic:

• Use of machine learning methods (deep learning, neural networks) for image analysis and extraction of relevant scene features.
• Integrate formal logic systems (predicate logic, mereological logic) to model relationships between objects and generate consistent situational descriptions.

Mereological methods in computer vision:

• Development and implementation of mereological methods in the context of vision systems and machine learning to enable formal representation of parts, wholes and their interrelationships in visual scenes.

Anticipated results

Anticipated results of the research program:

• Develop a vision system that generates logical, consistent situational descriptions for images and video sequences.
• Implementing mereological methods to analyze spatial relationships between objects in images.
• Combining semantic segmentation with NLP methods to automatically create scene descriptions.
• Scientific publications and implementations of prototype tools in the areas of computer vision, artificial intelligence and natural language processing.

Program innovation:

The program integrates approaches from computer vision, AI, logic and NLP in a comprehensive manner, emphasizing the generation of descriptions that are not only visually correct, but also logically consistent and semantically meaningful. The introduction of mereological methods will allow more accurate modeling of part-whole relationships in visual scenes, which is novel in the context of visual machine learning.

I invite you to cooperate with us!

Research Program Manager
Adam Szmigielski
aszmigie@pjwstk.edu.pl