S-Space College of Humanities (인문대학) Program in Cognitive Science (협동과정-인지과학전공) Theses (Master's Degree_협동과정-인지과학전공)
Investigating Evidence for Hierarchical Color Learning in Convolutional Neural Networks
콘볼루션 신경망의 색상 위계 학습에 대한 탐구
- Chris Hickey
- Issue Date
- 서울대학교 대학원
- color hierarchy; computer vision; convolutional neural network; cognitive universalities; faster R-CNN; 색상 위계; 컴퓨터 비전; 콘볼루션 신경망; 인지과정의 보편성; 물체 탐지
- 학위논문 (석사) -- 서울대학교 대학원 : 인문대학 협동과정 인지과학전공, 2020. 8. 장병탁.
- Empirical evidence suggests that color categories emerge in a universal, recurrent, hierarchical pattern across different cultures in the following order; white, black < red < green, yellow < blue < brown < pink, gray, orange, and purple}. This pattern is referred to as the "Color Hierarchy". Over two experiments, the present study examines whether there is evidence for such hierarchical color category learning patterns in Convolutional Neural Networks (CNNs). Experiment A investigates whether color categories are learned randomly, or in a fixed, hierarchical fashion. Results show that colors higher up the Color Hierarchy (e.g. red) are generally learned before colors lower down the hierarchy (e.g. brown, orange, gray). Experiment B examines whether object color affects recall in object detection. Similar to Experiment A, results show that object recall is noticeably impacted by color, with colors higher up the Color Hierarchy generally showing better recall. Additionally, objects whose color can be described by adjectives that emphasise colorfulness (e.g. vivid, brilliant, deep) show better recall than those which de-emphasise colorfulness (e.g. dark, pale, light). The effect of both color hue and adjective on object recall is still observable, even when controlling for contrast through grayscale images. These results highlight similarities between humans and CNNs in color perception, and provide insight into factors that influence object detection. They also show the value of using deep learning techniques as a means of investigating cognitive universalities in an efficient, unbiased, cost-effective way.
경험적으로 색상 계열은 보편적으로 다양한 문화권에 걸쳐 순환적이고 위계적인 패턴을 나타내며 그 순서는 다음과 같이 나타난다; 검은색 < 붉은색