模棱两可
感知
集合(抽象数据类型)
数学
代表(政治)
计算机科学
人工智能
球(数学)
心理学
认知心理学
几何学
政治学
政治
神经科学
程序设计语言
法学
作者
Maarten Wijntjes,Robert Volcic
出处
期刊:Journal of Vision
[Association for Research in Vision and Ophthalmology (ARVO)]
日期:2019-09-06
卷期号:19 (10): 243c-243c
摘要
Online shopping confronts us with an interesting problem: we first see the representation of something, and only later we see it in reality. If reality disappoints us, we return the product, which is an unsustainable scenario. This motivates the search for optimal (visual) representations that predict reality as good as possible. We investigated how to quantify the quality of apparel photos by measuring a set of perceived material attributes. We used images of 28 fabrics represented in four different styles (draped over ball, flat with wrinkles, 2 zoomed-in versions with different light). We let observers (n=153, distributed over different blocks) rate six attributes: warmth, softness, weight, shininess, elasticity and roughness on a continuous scale in an online experiment. Each setting was repeated 3 times. First, we computed intra-correlations of the 28 ratings for three repetition pairs and we found that they were stable across both attributes and photo styles. This implies that observers’ internal consistency did not depend on either attribute or style. Next, we analyzed correlations between observers, a measure of visual ambiguity. We found these to vary for both attributes and photo styles. Zoomed-in pictures were more ambiguous than zoomed-out. Furthermore, ‘shininess’ was least ambiguous, followed by ‘softness’, while the most ambiguous was ‘warmth’. We also performed a Principle Component Analysis which revealed that low dimensional embeddings had markedly different loadings for each or the four photo styles. Interestingly, the difference between the two zoomed-in picture was larger than between zoomed-out pictures. Moreover, explained variance was higher for first PCA components in the zoomed-out pictures. This results reveal that photographic styles influence perceived material qualities, and that certain styles are less ambiguous than others. Furthermore, this paradigm can easily be used for the ultimate test case of seeing and/or feeling the fabrics in reality.
科研通智能强力驱动
Strongly Powered by AbleSci AI