The capacity of Vision transformers (ViTs) to handle variable-sized inputs is often constrained by computational complexity and batch processing limitations. Consequently, ViTs are typically trained on small, fixed-size images obtained through downscaling or cropping. While reducing computational burden, these methods result in significant information loss, negatively affecting tasks like image aesthetic assessment. We introduce Charm, a novel tokenization approach that preserves Composition, High-resolution, Aspect Ratio, and Multi-scale information simultaneously. Charm prioritizes high-resolution details in specific regions while downscaling others, enabling shorter fixed-size input sequences for ViTs while incorporating essential information. Charm is designed to be compatible with pre-trained ViTs and their learned positional embeddings. By providing multiscale input and introducing variety to input tokens, Charm improves ViT performance and generalizability for image aesthetic assessment. We avoid cropping or changing the aspect ratio to further preserve information. Extensive experiments demonstrate significant performance improvements on various image aesthetic and quality assessment datasets (up to 8.1 %) using a lightweight ViT backbone. Code and pre-trained models are available at https://github.com/FBehrad/Charm.

Deep neural networks perform well in object recognition, but do they perceive objects like humans?This study investigates the Gestalt principle of closure in convolutional neural networks.We propose a protocol to identify closure and conduct experiments using simple visual stimuli with progressively removed edge sections.We evaluate well-known networks on their ability to classify incomplete polygons.Our findings reveal a performance degradation as the edge removal percentage increases, indicating that current models heavily rely on complete edge information for accurate classification.The data used in our study is available on Github.* This work was

Two stimuli alternately presented at different locations can evoke a percept of a stimulus continuously moving between the two locations. The neural mechanism underlying this apparent motion (AM) is thought to be increased activation of primary visual cortex (V1) neurons tuned to locations along the AM path, although evidence remains inconclusive. AM masking, which refers to the reduced detectability of stimuli along the AM path, has been taken as evidence for AM-related V1 activation. AM-induced neural responses are thought to interfere with responses to physical stimuli along the path and as such impair the perception of these stimuli. However, AM masking can also be explained by predictive coding models, predicting that responses to stimuli presented on the AM path are suppressed when they match the spatio-temporal prediction of a stimulus moving along the path. In the present study, we find that AM has a distinct effect on the detection of target gratings, limiting the maximum performance at high contrast levels. This masking is strongest when the target orientation is identical to the orientation of the inducers. We developed a V1-like population code model of early visual processing, based on a standard contrast normalization model. We find that AM-related activation in early visual cortex is too small to either cause masking or to be perceived as motion. Our model instead predicts strong suppression of early sensory responses during AM, consistent with the theoretical framework of predictive coding.

Several studies have reported optimal population decoding of sensory responses in two-alternative visual discrimination tasks. Such decoding involves integrating noisy neural responses into a more reliable representation of the likelihood that the stimuli under consideration evoked the observed responses. Importantly, an ideal observer must be able to evaluate likelihood with high precision and only consider the likelihood of the two relevant stimuli involved in the discrimination task. We report a new perceptual bias suggesting that observers read out the likelihood representation with remarkably low precision when discriminating grating spatial frequencies. Using spectrally filtered noise, we induced an asymmetry in the likelihood function of spatial frequency. This manipulation mainly affects the likelihood of spatial frequencies that are irrelevant to the task at hand. Nevertheless, we find a significant shift in perceived grating frequency, indicating that observers evaluate likelihoods of a broad range of irrelevant frequencies and discard prior knowledge of stimulus alternatives when performing two-alternative discrimination.

Human subjects easily perceive and extensively use shape regularities such as symmetry or periodicity when they are confronted with tile task of object description and recognition. A computer vision algorithm is presented that emulates such behavior in that it similarly makes use of shape redundancies for the concise description and meaningful segmentation of planar object contours. The contours are analyzed in so-called arc length space. This parameter space facilitates the detection of regularities under a broad range of viewing conditions. Several of the irregularities which have traditionally been treated in isolation, are given a unified substrate for their detection and use in building compact models. Regularity consistency checks are made and, if necessary, altered versions are inferred.>