A Pig, an Angel and a Cactus Walk Into a Blender
We implemented a system for automatic generation of visual blends (Blender) composed of two components: the Mapper and the Visual Blender (Fig. 1). We consider it a hybrid blender, as the blending process starts at the conceptual level (in the Mapper) and only ends at the visual representation level (in the Visual Blender). The focus is given to the Visual Blender (read more about it).
Structure of the implemented Blender. The Mapper receives two input spaces (1), one referring to concept A and the other one to concept B. It produces analogies (3) that are afterwards used by the Visual Blender component. The Visual Blender also receives visual representations and corresponding list of relations among parts (2) that are used as a base and data for producing the visual blends (4).
Collecting data
The goal of the initial phase of the project was to collect visual representations for three previously selected concepts: angel (human-like entity), pig (animal) and cactus (plant). The enquiry was composed of five tasks:
- Collection of visual representations for the selected concepts;
- Identification of the representational elements;
- Description of the relations among the identified elements;
- Identification of the prototypical elements;
- Collection of visual blends for the selected concepts.
The data was collected from nine participants and the representations for each of the concepts were converted into fully scalable vector graphics (Fig. 2) and prepared to be used as base visual representations for the Visual Blender.
Generating the blends
Having the organization of mental spaces as an inspiration, we follow a similar approach to structure the construction of the visual representations, which are considered as a group of several parts / elements. A visual representation for a given concept is constructed as a well-structured object which can contain other objects, storing the following attributes:
- Name, shape, position relative to the father-object;
- The set of relations to other objects. A relation between two objects consists of: object A, object B and type of relation – e.g. eye (A) inside head (B);
- The set of child-objects.
In the Visual Blender component we use an evolutionary engine in which each population corresponds to a different analogy (produced by the Mapper) and each individual is a visual blend. The evolution is guided by a fitness function that assesses the quality of each blend based on the satisfied relations.
Using this descriptive approach, the complexity of blending two base representations is reduced, as it facilitates object exchange and recursive changing (by moving an object, the child-objects are also easily moved). Additionally, it also contributes to the overall sense of cohesion among the parts and gives much more flexibility to the construction of representations by allowing the generation of unlimited number of similar (and also valid) blends, based on the same set of relations.
Examples of produced blends.
The Results
Overall, the analysis of the experimental results indicates that the implemented blender is able to produce sets of blends with great variability and unexpected features, while respecting the analogy. The evolutionary engine is capable of evolving the blends towards a higher number of satisfied relations.
By comparing the produced blends with the ones drawn by the participants, we can say that the Blender is not only able to produce blends that are coherent with the human-drawn ones (Fig. 4) but is also able to produce novel blends that no participant drew (Fig. 3).
Comparison between hand-drawn blends and blends generated by the implemented Blender, organised by groups: group (1) corresponds to pig-cactus blends; (2) corresponds to angel-cactus; groups (3-5) correspond to pig- angel (the figure on the left of each group is the hand-drawn blend).
In order to assess if the produced blends could be correctly perceived, a second enquiry was conducted. The main goal was to evaluate whether or not the participant could identify the input spaces used for each blend (i.e. if it was possible to identify pig and cactus in a blend produced for pig-cactus). Overall, the majority of the participants could identify at least one of the input spaces for the “good” exemplars of visual blends (exemplars with the prototypical parts clearly identifiable). Even though some of the participants could not correctly name both of the input spaces, the answers given were somehow related to the correct ones.
Publications
-
J. M. Cunha, J. Gonçalves, P. Martins, P. Machado, and A. Cardoso, “A Pig, an Angel and a Cactus Walk Into a Blender: A Descriptive Approach to Visual Blending,” in Proceedings of the Eighth International Conference on Computational Creativity (ICCC 2017), 2017.
- Bibtex
- |
@inproceedings{cgmmc2017,
author = {Cunha, Jo\~{a}o Miguel and Gon\c{c}alves, Jo\~{a}o and Martins, Pedro and Machado, Penousal and Cardoso, Am\'{i}lcar},
Booktitle = {Proceedings of the Eighth International Conference on Computational Creativity (ICCC 2017)},
Title = {A Pig, an Angel and a Cactus Walk Into a Blender: A Descriptive Approach to Visual Blending},
Year = {2017},
}