In visual occlusion 2 amodal-completion tendencies occur frequently. One tendency leads toward the simplest completed shape (a global completion) and the other to a shape for which the completion itself is as simple as possible (a local completion). Two experimental paradigms were used to test the strengths of these completion tendencies: a drawing task and a simultaneous matching task. The experimental results support the notion that the preference for either a global or a local completion is the consequence of a competition between interpretations. Finally, the authors discuss how the preference for a completion can be predicted by a model that is based on a quantification of both global and local aspects.

Explainability and interpretability are two critical aspects of decision support systems. Within computer vision, they are critical in certain tasks related to human behavior analysis such as in health care applications. Despite their importance, it is only recently that researchers are starting to explore these aspects. This paper provides an introduction to explainability and interpretability in the context of computer vision with an emphasis on looking at people tasks. Specifically, we review and study those mechanisms in the context of first impressions analysis. To the best of our knowledge, this is the first effort in this direction. Additionally, we describe a challenge we organized on explainability in first impressions analysis from video. We analyze in detail the newly introduced data set, evaluation protocol, proposed solutions and summarize the results of the challenge. Finally, derived from our study, we outline research opportunities that we foresee will be decisive in the near future for the development of the explainable computer vision field.Keywords Explainable computer vision · First impressions · Personality analysis · Multimodal information · Algorithmic accountability 1 IntroductionLooking at People (LaP) -the field of research focused on the visual analysis of human behavior -has been a very active research field within computer vision in the last decade [28,29,62]. Initially, LaP focused on tasks associated with basic human behaviors that were obviously visual (e.g., basic gesture recognition [71,70] or face recognition in restricted scenarios [10,83]). Research progress in LaP has now led to models that can solve those initial tasks relatively easily [66,82]. Instead, attention on human behavior analysis has now turned to problems that are not visually evident to model / recognize [84,48,72]. For instance, consider the task of assessing personality traits from visual information [72]. Although there are methods that can estimate apparent personality traits with (relatively) acceptable performance, model recommendations by themselves are useless if the end user is not confident on the model's reasoning, as the primary use for such estimation is to understand bias in human assessors.Explainability and interpretability are thus critical features of decision support systems in some LaP tasks [26]. The former focuses on mechanisms that can tell what is the rationale behind the decision or recommendation made by

The topic of amodal completion has often been investigated by using partly occluded shapes that are regular. In research that has typically been done with displays such as these regular shapes, it has been shown that global aspects of a shape can determine completion. To see how robust these global influences in the completion process are, we investigated quasi-regular shapes, ie shapes with a certain overall regularity but not based on metrical identities. First, in experiment 1 participants had to complete quasi-regular shapes in a drawing task. Then, in experiment 2 the primed-matching paradigm was used. Results from both experiments provided evidence for global completions. In experiment 3 we found that multiple global completions can be primed, which, as a control experiment showed, cannot be explained by some inability of the visual system to see the difference between the different completions. These data support the notion that global influences on visual occlusion are apparent even when the partly occluded stimulus is outside the domain of regular shapes. Implications for a global approach are provided.

We studied the effects of learning on amodal completion of partly occluded shapes. Amodal completion may originate from local characteristics of the partly occluded contours, resulting in local completions, or from global characteristics, resulting in global completions. Two classes of occlusion patterns were constructed: convergent occlusion patterns, in which global and local completions resulted in the same shape, and the much more ambiguous divergent occlusion patterns, in which these completions resulted in different shapes. We used a sequential matching paradigm and obtained behavioral responses (Experiment 1s and 2) and electroencephalogram recordings (Experiment 3) to investigate whether previously learned shapes influenced completions of partly occluded shapes. Experiment 1 revealed the preference for different completions of both occlusion patterns. In Experiment 2, learning effects were found only for test shapes following divergent occlusion patterns. Experiment 3 showed differential effects with regard to convergent and divergent occlusion patterns on a positive event-related potential in the 150- to 300-ms range, before learning. After learning, modulation of this effect was only found for the divergent occlusion patterns. The results show that amodal completion of shapes can be influenced by a simple learning task when multiple completions of partly occluded shapes are perceptually plausible.

In perception research, various models have been designed for the encoding of, for example. visual patterns, in order to predict the human interpretation of such patterns. Each of these encoding models provides a few coding rules to obtain codes for a pattern, each code expressing regularity and hierarchy in that pattern. Some of these models employ the minimum principle which states that the human interpretation of a pattern is reflected by the simplest code for that pattern. ie the simplest code according to a given complexitv metric. In rhis paper a new complexitv metric is proposed. This metric is based on a formal analysis of the concept of regularitv. Some conclusions of this analysis are sketched. The new metric does not depend on artifacts of the coding rules. [t accounts for the amounts of irregularity and hierarchy as represented in a code of a pattern. such that these two amounts can be added to determine the complexity of a code. An experiment is discussed that shows that the new metric performs significantly better than the metrics used previously. In particular, the new metric predicts more local pattern organizations than the old metrics. This implies that various local pattern organizations do not falsif-v the minimum prirrciple anymore.