|
|
Procedural Modeling of Tree Bark
Laitoch, Petr ; Beneš, Jan (advisor) ; Kahoun, Martin (referee)
Even though procedural modeling of trees is a well-studied problem, realistic modeling of tree bark is not. However, the more general techniques of texture syn- thesis, including texture-by-numbers, may be helpful for modeling tree bark. Tex- ture synthesis is a process of generating an arbitrarily large texture similar to an input image. This method is capable of generating homogeneous textures. This is not enough as many types of bark are inhomogeneous. Texture-by-numbers improves texture synthesis by further guiding the process with provided label maps to allow the generation of even inhomogeneous textures. Many texture-by- numbers algorithms are not currently implemented. In this thesis, we implement a promising texture-by-numbers algorithm along with algorithms for generating the required label maps. This combination of algorithms creates a pipeline for synthesizing realistic tree bark textures based on a single small input image. We test out the pipeline on samples of multiple types of real-world tree bark images and discuss the results. We further suggest multiple directions for improving the employed techniques.
|
|
|
Text classification with limited training data
Laitoch, Petr ; Hana, Jiří (advisor) ; Vidová Hladká, Barbora (referee)
The aim of this thesis is to minimize manual work needed to create training data for text classification tasks. Various research areas including weak supervision, interactive learning and transfer learning explore how to minimize training data creation effort. We combine ideas from available literature in order to design a comprehensive text classification framework that employs keyword-based labeling instead of traditional text annotation. Keyword-based labeling aims to label texts based on keywords contained in the texts that are highly correlated with individual classification labels. As noted repeatedly in previous work, coming up with many new keywords is challenging for humans. To accommodate for this issue, we propose an interactive keyword labeler featuring the use of word similarity for guiding a user in keyword labeling. To verify the effectiveness of our novel approach, we implement a minimum viable prototype of the designed framework and use it to perform a user study on a restaurant review multi-label classification problem.
|
|
|
Procedural Modeling of Tree Bark
Laitoch, Petr ; Beneš, Jan (advisor) ; Kahoun, Martin (referee)
Even though procedural modeling of trees is a well-studied problem, realistic modeling of tree bark is not. However, the more general techniques of texture syn- thesis, including texture-by-numbers, may be helpful for modeling tree bark. Tex- ture synthesis is a process of generating an arbitrarily large texture similar to an input image. This method is capable of generating homogeneous textures. This is not enough as many types of bark are inhomogeneous. Texture-by-numbers improves texture synthesis by further guiding the process with provided label maps to allow the generation of even inhomogeneous textures. Many texture-by- numbers algorithms are not currently implemented. In this thesis, we implement a promising texture-by-numbers algorithm along with algorithms for generating the required label maps. This combination of algorithms creates a pipeline for synthesizing realistic tree bark textures based on a single small input image. We test out the pipeline on samples of multiple types of real-world tree bark images and discuss the results. We further suggest multiple directions for improving the employed techniques.
|
guest ::
