Search

People see some impossible events as more impossible than others. For example, walking through a solid wall seems more impossible if it is made of stone rather than wood. Across four experiments, we investigated how children and adults assess the relative impossibility of events, contrasting two kinds of information they may use: perceptual information and causal knowledge. In each experiment, participants were told about a wizard who could magically transform target objects into other things. Participants then assessed which of the two transformation spells would be easier or harder, a spell transforming a target object into a perceptual match (i.e., a similar-looking thing) or one transforming it into a causal match (e.g., an item made of similar materials). In Experiments 1–3, children aged 4–7 mainly thought that transformations into the perceptual match would be easier, though this tendency varied with age. Adults were overall split when choosing which spell would be easier. In Experiment 1, this was because of variations in their judgments across different pairs of spells; in Experiments 2 and 4, the split resulted because different subsets of adults preferred either the perceptual or causal match. Overall, these findings show that children, and many adults, use perceptual reasoning to assess relative impossibility. Public Significance Statement: Besides differentiating possible events from ones that are impossible, people see some events as more impossible than others. We investigated how children and adults make these judgments by asking them about magical transformations. One way that people might assess the relative impossibility of magic is by drawing on their causal knowledge. However, we find that young children, and even many adults, instead make these judgments using shallow perceptual information. [ABSTRACT FROM AUTHOR]

Accurate and precise rigid registration between head-neck computed tomography (CT) and cone-beam computed tomography (CBCT) images is crucial for correcting setup errors in image-guided radiotherapy (IGRT) for head and neck tumors. However, conventional registration methods that treat the head and neck as a single entity may not achieve the necessary accuracy for the head region, which is particularly sensitive to radiation in radiotherapy. We propose ACSwinNet, a deep learning-based method for head-neck CT-CBCT rigid registration, which aims to enhance the registration precision in the head region. Our approach integrates an anatomical constraint encoder with anatomical segmentations of tissues and organs to enhance the accuracy of rigid registration in the head region. We also employ a Swin Transformer-based network for registration in cases with large initial misalignment and a perceptual similarity metric network to address intensity discrepancies and artifacts between the CT and CBCT images. We validate the proposed method using a head-neck CT-CBCT dataset acquired from clinical patients. Compared with the conventional rigid method, our method exhibits lower target registration error (TRE) for landmarks in the head region (reduced from 2.14 ± 0.45 mm to 1.82 ± 0.39 mm), higher dice similarity coefficient (DSC) (increased from 0.743 ± 0.051 to 0.755 ± 0.053), and higher structural similarity index (increased from 0.854 ± 0.044 to 0.870 ± 0.043). Our proposed method effectively addresses the challenge of low registration accuracy in the head region, which has been a limitation of conventional methods. This demonstrates significant potential in improving the accuracy of IGRT for head and neck tumors. [ABSTRACT FROM AUTHOR]

Accurately detecting defects while reconstructing a high-quality normal background in surface defect detection using unsupervised methods remains a significant challenge. This study proposes an unsupervised method that effectively addresses this challenge by achieving both accurate defect detection and a high-quality normal background reconstruction without noise. We propose an adaptive weighted structural similarity (AW-SSIM) loss for focused feature learning. AW-SSIM improves structural similarity (SSIM) loss by assigning different weights to its sub-functions of luminance, contrast, and structure based on their relative importance for a specific training sample. Moreover, it dynamically adjusts the Gaussian window's standard deviation (σ) during loss calculation to balance noise reduction and detail preservation. An artificial defect generation algorithm (ADGA) is proposed to generate an artificial defect closely resembling real ones. We use a two-stage training strategy. In the first stage, the model trains only on normal samples using AW-SSIM loss, allowing it to learn robust representations of normal features. In the second stage of training, the weights obtained from the first stage are used to train the model on both normal and artificially defective training samples. Additionally, the second stage employs a combined learned Perceptual Image Patch Similarity (LPIPS) and AW-SSIM loss. The combined loss helps the model in achieving high-quality normal background reconstruction while maintaining accurate defect detection. Extensive experimental results demonstrate that our proposed method achieves a state-of-the-art defect detection accuracy. The proposed method achieved an average area under the receiver operating characteristic curve (AuROC) of 97.69% on six samples from the MVTec anomaly detection dataset. [ABSTRACT FROM AUTHOR]

Across the millennia, and across a range of disciplines, there has been a widespread desire to connect, or translate between, the senses in a manner that is meaningful, rather than arbitrary. Early examples were often inspired by the vivid, yet mostly idiosyncratic, crossmodal matches expressed by synaesthetes, often exploited for aesthetic purposes by writers, artists, and composers. A separate approach comes from those academic commentators who have attempted to translate between structurally similar dimensions of perceptual experience (such as pitch and colour). However, neither approach has succeeded in delivering consensually agreed crossmodal matches. As such, an alternative approach to sensory translation is needed. In this narrative historical review, focusing on the translation between audition and vision, we attempt to shed light on the topic by addressing the following three questions: (1) How is the topic of sensory translation related to synaesthesia, multisensory integration, and crossmodal associations? (2) Are there common processing mechanisms across the senses that can help to guarantee the success of sensory translation, or, rather, is mapping among the senses mediated by allegedly universal (e.g., amodal) stimulus dimensions? (3) Is the term 'translation' in the context of cross-sensory mappings used metaphorically or literally? Given the general mechanisms and concepts discussed throughout the review, the answers we come to regarding the nature of audio-visual translation are likely to apply to the translation between other perhaps less-frequently studied modality pairings as well. [ABSTRACT FROM AUTHOR]

Considering working memory capacity limitations, representing all relevant data simultaneously is unlikely. What remains unclear is why some items are better remembered than others when all data are equally relevant. While trying to answer this question, the literature has identified a pattern named the mixed-category benefit in which performance is enhanced when presenting stimuli from different categories as compared to presenting a similar number of items that all belong to just one category. Moreover, previous studies revealed an asymmetry in performance while mixing certain categories, suggesting that not all categories benefit equally from being mixed. In a series of three change-detection experiments, the present study investigated the role of low-level perceptual similarities between categories in determining the mixed-category asymmetric advantages. Our primary conclusion is that items' similarity at the perceptual level has a significant role in the asymmetric performance in the mixed-category phenomenon. We measured sensitivity (d′) to detect a change between sample and test displays and found that the mixed-category advantage dropped when the mixed categories shared basic features. Furthermore, we found that sensitivity to novel items was impaired when presented with another category sharing its basic features. Finally, increasing the encoding interval improved performance for the novel items, but novel items' performance was still impaired when these items were mixed with another category that shared their basic features. Our findings highlight the significant role low-level similarities play in the asymmetric mixed-category performances, for both novel and familiar categories. Public Significance Statement: Our visual environment is a diverse collection of stimuli sourced from various categories. Research has shown that memory performance is better when different categories of stimuli are combined, rather than when a single category is presented. However, recent studies have revealed that not all combinations of stimuli produce comparable memory enhancement, which has resulted in an unresolved asymmetry in the mixed-category advantage. Our findings, in contrast to previous interpretations that primarily rely on high-level cognitive processing, demonstrate that low-level encoding processes also impact memory enhancement. Specifically, our study provides evidence that the perceptual similarities between stimuli derived from different categories have an effect on the extent of the mixed-category advantage. [ABSTRACT FROM AUTHOR]

Accurate and precise rigid registration between head-neck computed tomography (CT) and cone-beam computed tomography (CBCT) images is crucial for correcting setup errors in image-guided radiotherapy (IGRT) for head and neck tumors. However, conventional registration methods that treat the head and neck as a single entity may not achieve the necessary accuracy for the head region, which is particularly sensitive to radiation in radiotherapy. We propose ACSwinNet, a deep learning-based method for head-neck CT-CBCT rigid registration, which aims to enhance the registration precision in the head region. Our approach integrates an anatomical constraint encoder with anatomical segmentations of tissues and organs to enhance the accuracy of rigid registration in the head region. We also employ a Swin Transformer-based network for registration in cases with large initial misalignment and a perceptual similarity metric network to address intensity discrepancies and artifacts between the CT and CBCT images. We validate the proposed method using a head-neck CT-CBCT dataset acquired from clinical patients. Compared with the conventional rigid method, our method exhibits lower target registration error (TRE) for landmarks in the head region (reduced from 2.14 ± 0.45 mm to 1.82 ± 0.39 mm), higher dice similarity coefficient (DSC) (increased from 0.743 ± 0.051 to 0.755 ± 0.053), and higher structural similarity index (increased from 0.854 ± 0.044 to 0.870 ± 0.043). Our proposed method effectively addresses the challenge of low registration accuracy in the head region, which has been a limitation of conventional methods. This demonstrates significant potential in improving the accuracy of IGRT for head and neck tumors.

Training well-performing Generative Adversarial Networks (GANs) with limited data has always been challenging. Existing methods either require sufficient data (over 100 training images) for training or generate images of low quality and low diversity. To solve this problem, we propose a new Cross-domain Semantic Relation (CSR) loss. The CSR loss improves the performance of the generative model by maintaining the relationship between instances in the source domain and generated images. At the same time, a perceptual similarity loss and a discriminative contrastive loss are designed to further enrich the diversity of generated images and stabilize the training process of models. Experiments on nine publicly available few-shot datasets and comparisons with the current nine methods show that our approach is superior to all baseline methods. Finally, we perform ablation studies on the proposed three loss functions and prove that these three loss functions are essential for few-shot image generation tasks. Code is available at https://github.com/gouayao/CSR. [ABSTRACT FROM AUTHOR]

Real-time estimation of actual object depth is an essential module for various autonomous system tasks such as 3D reconstruction, scene understanding and condition assessment. During the last decade of machine learning, extensive deployment of deep learning methods to computer vision tasks has yielded approaches that succeed in achieving realistic depth synthesis out of a simple RGB modality. Most of these models are based on paired RGB-depth data and/or the availability of video sequences and stereo images. However, the lack of RGB-depth pairs, video sequences, or stereo images makes depth estimation a challenging task that needs to be explored in more detail. This study builds on recent advances in the field of generative neural networks in order to establish fully unsupervised single-shot depth estimation. Two generators for RGB-to-depth and depth-to-RGB transfer are implemented and simultaneously optimized using the Wasserstein-1 distance, a novel perceptual reconstruction term, and hand-crafted image filters. We comprehensively evaluate the models using a custom-generated industrial surface depth data set as well as the Texas 3D Face Recognition Database, the CelebAMask-HQ database of human portraits and the SURREAL dataset that records body depth. For each evaluation dataset, the proposed method shows a significant increase in depth accuracy compared to state-of-the-art single-image transfer methods. [ABSTRACT FROM AUTHOR]

This study aims to demonstrate that the adaptation of loanwords cannot be solely explained by speech perception. Specifically, it focuses on the adaptation of the English /s/ sound into Korean. In English, a singleton /s/ is loaned as tense [s*], while a preconsonantal /s/ is adapted as lenis [s] in Korean. Kim and Curtis (2002) and Kang (2008) provided an acoustic cue-based analysis for the adaptation of English /s/, considering duration and voice quality, respectively. To investigate the perception of English /s/ by Korean listeners, we conducted a perception experiment. We examined the impact of prosodic factors, such as domain-edge position and lexical stress, on the perception of /s/. Our findings revealed diverse perception patterns depending on stress and position within a word. Specifically, /s/ in a stressed syllable and /s/ in a word-edge position were more likely to be perceived as tense [s*]. Contrary to Kim and Curtis (2002), the duration of English /s/ alone was insufficient to explain perception patterns. Based on our varied perception patterns, we argue that categorical adaptation of the singleton /s/ as tense [s*] and the preconsonantal /s/ as lenis [s] does not solely reflect speech perception. Instead, we propose that loanword adaptation is a process driven by top-down/phonotactic expectations and acoustic matching, with the goal of maximizing perceptual similarity between the source word and the loanword. This is consistent with the findings of Daland et al. (2019). [ABSTRACT FROM AUTHOR]

Accurately detecting defects while reconstructing a high-quality normal background in surface defect detection using unsupervised methods remains a significant challenge. This study proposes an unsupervised method that effectively addresses this challenge by achieving both accurate defect detection and a high-quality normal background reconstruction without noise. We propose an adaptive weighted structural similarity (AW-SSIM) loss for focused feature learning. AW-SSIM improves structural similarity (SSIM) loss by assigning different weights to its sub-functions of luminance, contrast, and structure based on their relative importance for a specific training sample. Moreover, it dynamically adjusts the Gaussian window’s standard deviation (σ) during loss calculation to balance noise reduction and detail preservation. An artificial defect generation algorithm (ADGA) is proposed to generate an artificial defect closely resembling real ones. We use a two-stage training strategy. In the first stage, the model trains only on normal samples using AW-SSIM loss, allowing it to learn robust representations of normal features. In the second stage of training, the weights obtained from the first stage are used to train the model on both normal and artificially defective training samples. Additionally, the second stage employs a combined learned Perceptual Image Patch Similarity (LPIPS) and AW-SSIM loss. The combined loss helps the model in achieving high-quality normal background reconstruction while maintaining accurate defect detection. Extensive experimental results demonstrate that our proposed method achieves a state-of-the-art defect detection accuracy. The proposed method achieved an average area under the receiver operating characteristic curve (AuROC) of 97.69% on six samples from the MVTec anomaly detection dataset.

Simple Summary: The impact of basal cell carcinomas (BCCs) on a patient's appearance can be significant. Reliable assessments are crucial for the effective management and evaluation of therapeutic interventions. Given that color and texture are critical attributes that describe the clinical aspect of skin lesions, our focus was to devise metrics that capture the way experts perceive deviations of target BCC areas from the surrounding healthy skin. Using computerized image analysis, we explored various similarity metrics to predict perceptual similarity, including different color spaces and distances between features from image embeddings derived from a pre-trained deep convolutional neural network. The results are promising in providing a valid, reliable, and affordable modality, enabling more accurate and standardized assessments of BCC tumors and post-treatment scars. Our approach to modeling color and texture lesion similarity from the surrounding healthy skin is a promising paradigm for the further development of a valid and reliable scar assessment tool. Efficient management of basal cell carcinomas (BCC) requires reliable assessments of both tumors and post-treatment scars. We aimed to estimate image similarity metrics that account for BCC's perceptual color and texture deviation from perilesional skin. In total, 176 clinical photographs of BCC were assessed by six physicians using a visual deviation scale. Internal consistency and inter-rater agreement were estimated using Cronbach's α, weighted Gwet's AC2, and quadratic Cohen's kappa. The mean visual scores were used to validate a range of similarity metrics employing different color spaces, distances, and image embeddings from a pre-trained VGG16 neural network. The calculated similarities were transformed into discrete values using ordinal logistic regression models. The Bray–Curtis distance in the YIQ color model and rectified embeddings from the 'fc6' layer minimized the mean squared error and demonstrated strong performance in representing perceptual similarities. Box plot analysis and the Wilcoxon rank-sum test were used to visualize and compare the levels of agreement, conducted on a random validation round between the two groups: 'Human–System' and 'Human–Human.' The proposed metrics were comparable in terms of internal consistency and agreement with human raters. The findings suggest that the proposed metrics offer a robust and cost-effective approach to monitoring BCC treatment outcomes in clinical settings. [ABSTRACT FROM AUTHOR]

Visual recognition requires inferring the similarity between a perceived object and a mental target. However, a measure of similarity is difficult to determine when it comes to complex stimuli such as faces. Indeed, people may notice someone "looks like" a familiar face, but find it hard to describe on the basis of what features such a comparison is based. Previous work shows that the number of similar visual elements between a face pictogram and a memorized target correlates with the P300 amplitude in the visual evoked potential. Here, we redefine similarity as the distance inferred from a latent space learned using a state‐of‐the‐art generative adversarial neural network (GAN). A rapid serial visual presentation experiment was conducted with oddball images generated at varying distances from the target to determine how P300 amplitude related to GAN‐derived distances. The results showed that distance‐to‐target was monotonically related to the P300, showing perceptual identification was associated with smooth, drifting image similarity. Furthermore, regression modeling indicated that while the P3a and P3b sub‐components had distinct responses in location, time, and amplitude, they were similarly related to target distance. The work demonstrates that the P300 indexes the distance between perceived and target image in smooth, natural, and complex visual stimuli and shows that GANs present a novel modeling methodology for studying the relationships between stimuli, perception, and recognition. [ABSTRACT FROM AUTHOR]

Mandarin speakers tend to adapt intervocalic nasals as either an onset of the following syllable (e.g. Bruno → bù.lŭ .n uò), as a nasal geminate (e.g. Daniel → dā n.n í.ěr), or as one of the above forms (e.g. Tiffany → dì.fú. n í or dì.fē n.n í). Huang and Lin (2013, 2016) identified two factors that may induce the nasal gemination repair: (1) when stress falls on the pre-nasal vowel and (2) when the pre-nasal vowel is a non-high lax vowel. They hypothesized that Mandarin Chinese speakers insert a nasal coda to perceptually approximate the stronger nasalization and longer syllable duration associated with the stressed syllables, and the shorter vowel duration of a lax vowel because the vowels in closed syllables are shorter in Mandarin. The results from two forced-choice identification experiments and an open-ended transcription task showed that although Mandarin speakers' choices of different repairs were indeed biased by the different phonetic manipulations, suggesting an effect of perceptual similarity, their decisions were mainly guided by native phonotactics. The overall findings suggest that phonotactic, phonetic, as well as non-linguistic (i.e. frequency) factors interact with each other, resulting in the variable adaptation pattern. [ABSTRACT FROM AUTHOR]

In loanword phonology, perceptual similarity and faithful mapping of phonological contrast are two main factors which influence loanword adaptation. Previous studies observe that English phonological voicing contrast is mapped to Mandarin Chinese (hereafter, Mandarin) phonological aspiration contrast, indicating faithful mapping of phonological contrast. Nevertheless, the role of perceptual similarity in adaptation of English stops to Mandarin has not been fully explored. The current study investigates the influence of perceptual similarity on loanword adaptation by examining how English voiced and voiceless stops are adapted in Mandarin Chinese using a data set of 1427 novel Mandarin loanwords from English. The results show consistent adaptation of English voiced stops as Mandarin unaspirated stops and English aspirated voiceless stops as Mandarin aspirated ones, while inconsistent adaptation patterns are found for the English unaspirated voiceless stops. In particular, English post-/s/unaspirated voiceless stops which occupy a similar voice onset time (VOT) region to Mandarin unaspirated stops are adapted as Mandarin unaspirated ones, whereas the rest are mapped to aspirated stops in Mandarin. The overall adaptation patterns provide partial support to faithful mapping of phonological contrast and provide robust evidence for an integration of perceptual similarity with faithful mapping of phonological contrast in loanword adaptation. [ABSTRACT FROM AUTHOR]

Background The perceptual similarity between two objects, specifically similarity in the shape of the referents, is a crucial element for relating words in earlier stages of development. The role of this perceptual similarity has not been systematically explored in children with Down syndrome (DS). Method: The aim was to explore the role of perceptual similarity in relationships between words in children with DS. Two groups, children with typical development (TD) and children with DS, matched by gender and mental age, participated in a priming task with a preferential looking paradigm. The task presented validated perceptually-related word pairs (prime-target) and perceptually unrelated pairs. In the priming task both groups were asked to look at a target image (e.g. ball) that was perceptually related (e.g. cookie) or unrelated (e.g. skirt) to the prime. Results: Participants from both groups looked more at targets without perceptual similarity than at those with similarity to the prime, suggesting an inhibition effect. Conclusions: This finding suggests the role of visual information, particularly the shape of the referents, in the construction of the lexical system. [ABSTRACT FROM AUTHOR]

This paper explores the adaptation of English voiced fricative [z]E into Mandarin. The principal findings are that the adaptation of [z]E depends on its position in the source word. If [z]E occupies the word initial or middle position, it tends to be borrowed as Mandarin [ts]M or [s]M. If [z]E is in the ending position, more variations would be observed as it may correspond to [ts]M, [s]M, [ʈʂ]M or [ʂ]M. Moreover, adapting [z]E into Mandarin is also heavily influenced by the orthography. The English [z]E can be spelled with or . The letters of and are also listed in Mandarin Pinyin, a Romanization system that transcribes the sounds of Chinese characters using Roman alphabets. Specifically, is pronounced as [s]M, while [z] is pronounced as [ts]M. It turns out that [s]M sounds different from [z]E, while [ts]M is perceptually similar to [z]E. Generally, there emerged two major adaptation patterns, that is, [z]E  [ts]M and [z]E  [s]M. The latter is based on the spelling similarity, rather than perceptual similarity. It is therefore concluded that the loanword adaptation is not only determined by the speech perception, phonology and legitimacy of sound structures, but also systematically interfered with the orthography, that is, the source-loan spelling similarity. [ABSTRACT FROM AUTHOR]