Abstract: The advent of recurrent neural networks for handwriting recognition marked an important milestone reaching impressive recognition accuracies despite the great variability that we observe across different writing styles. Sequential architectures are a perfect fit to model text lines, not only because of the inherent temporal aspect of text, but also to learn probability distributions over sequences of characters and words. However, using such recurrent paradigms comes at a cost at training stage, since their sequential pipelines prevent parallelization. In this work, we introduce a non-recurrent approach to recognize handwritten text by the use of transformer models. We propose a novel method that bypasses any recurrence. By using multi-head self-attention layers both at the visual and textual stages, we are able to tackle character recognition as well as to learn language-related dependencies of the character sequences to be decoded. Our model is unconstrained to any predefined vocabulary, being able to recognize out-of-vocabulary words, i.e. words that do not appear in the training vocabulary. We significantly advance over prior art and demonstrate that satisfactory recognition accuracies are yielded even in few-shot learning scenarios.

Abstract: Effective conservation actions require effective population monitoring. However, accurately counting animals in the wild to inform conservation decision-making is difficult. Monitoring populations through image sampling has made data collection cheaper, wide-reaching and less intrusive but created a need to process and analyse this data efficiently. Counting animals from such data is challenging, particularly when densely packed in noisy images. Attempting this manually is slow and expensive, while traditional computer vision methods are limited in their generalisability. Deep learning is the state-of-the-art method for many computer vision tasks, but it has yet to be properly explored to count animals. To this end, we employ deep learning, with a density-based regression approach, to count fish in low-resolution sonar images. We introduce a large dataset of sonar videos, deployed to record wild Lebranche mullet schools (Mugil liza), with a subset of 500 labelled images. We utilise abundant unlabelled data in a self-supervised task to improve the supervised counting task. For the first time in this context, by introducing uncertainty quantification, we improve model training and provide an accompanying measure of prediction uncertainty for more informed biological decision-making. Finally, we demonstrate the generalisability of our proposed counting framework through testing it on a recent benchmark dataset of high-resolution annotated underwater images from varying habitats (DeepFish). From experiments on both contrasting datasets, we demonstrate our network outperforms the few other deep learning models implemented for solving this task. By providing an open-source framework along with training data, our study puts forth an efficient deep learning template for crowd counting aquatic animals thereby contributing effective methods to assess natural populations from the ever-increasing visual data.

Abstract: Lateral connections in the primary visual cortex (V1) have long been hypothesized to be responsible for several visual processing mechanisms such as brightness induction, chromatic induction, visual discomfort, and bottom-up visual attention (also named saliency). Many computational models have been developed to independently predict these and other visual processes, but no computational model has been able to reproduce all of them simultaneously. In this work, we show that a biologically plausible computational model of lateral interactions of V1 is able to simultaneously predict saliency and all the aforementioned visual processes. Our model's architecture (NSWAM) is based on Penacchio's neurodynamic model of lateral connections of V1. It is defined as a network of firing rate neurons, sensitive to visual features such as brightness, color, orientation, and scale. We tested NSWAM saliency predictions using images from several eye tracking data sets. We show that the accuracy of predictions obtained by our architecture, using shuffled metrics, is similar to other state-of-the-art computational methods, particularly with synthetic images (CAT2000-Pattern and SID4VAM) that mainly contain low-level features. Moreover, we outperform other biologically inspired saliency models that are specifically designed to exclusively reproduce saliency. We show that our biologically plausible model of lateral connections can simultaneously explain different visual processes present in V1 (without applying any type of training or optimization and keeping the same parameterization for all the visual processes). This can be useful for the definition of a unified architecture of the primary visual cortex.

Abstract: Rain removal plays an important role in the restoration of degraded images. Recently, CNN-based methods have achieved remarkable success. However, these approaches neglect that the appearance of real-world rain is often accompanied by low light conditions, which will further degrade the image quality, thereby hindering the restoration mission. Therefore, it is very indispensable to jointly remove the rain and enhance illumination for real-world rain image restoration. To this end, we proposed a novel spatially-adaptive network, dubbed SANet, which can remove the rain and enhance illumination in one go with the guidance of degradation mask. Meanwhile, to fully utilize negative samples, a contrastive loss is proposed to preserve more natural textures and consistent illumination. In addition, we present a new synthetic dataset, named DarkRain, to boost the development of rain image restoration algorithms in practical scenarios. DarkRain not only contains different degrees of rain, but also considers different lighting conditions, and more realistically simulates real-world rainfall scenarios. SANet is extensively evaluated on the proposed dataset and attains new state-of-the-art performance against other combining methods. Moreover, after a simple transformation, our SANet surpasses existing the state-of-the-art algorithms in both rain removal and low-light image enhancement.

Abstract: Visual illusions expand our understanding of the visual system by imposing constraints in the models in two different ways: i) visual illusions for humans should induce equivalent illusions in the model, and ii) illusions synthesized from the model should be compelling for human viewers too. These constraints are alternative strategies to find good vision models. Following the first research strategy, recent studies have shown that artificial neural network architectures also have human-like illusory percepts when stimulated with classical hand-crafted stimuli designed to fool humans. In this work we focus on the second (less explored) strategy: we propose a framework to synthesize new visual illusions using the optimization abilities of current automatic differentiation techniques. The proposed framework can be used with classical vision models as well as with more recent artificial neural network architectures. This framework, validated by psychophysical experiments, can be used to study the difference between a vision model and the actual human perception and to optimize the vision model to decrease this difference.

Abstract: Supervised training of neural networks requires a large amount of manually annotated data and the resulting networks tend to be sensitive to out-of-distribution (OOD) data.
Self- and semi-supervised training schemes reduce the amount of annotated data required during the training process. However, OOD generalization remains a major challenge for most methods. Strategies that promote smoother decision boundaries play an important role in out-of-distribution generalization. For example, embedding propagation (EP) for manifold smoothing has recently shown to considerably improve the OOD performance for few-shot classification. EP achieves smoother class manifolds by building a graph from sample embeddings and propagating information through the nodes in an unsupervised manner. In this work, we extend the original EP paper providing additional evidence and experiments showing that it attains smoother class embedding manifolds and improves results in settings beyond few-shot classification. Concretely, we show that EP improves the robustness of neural networks against multiple adversarial attacks as well as semi- and
self-supervised learning performance.

Abstract: This paper presents an algorithm for automated detection and classification of multi-cell phytoliths, one of the major components of many archaeological and paleoenvironmental deposits. This identification, based on phytolith wave pattern, is made using a pretrained VGG19 deep learning model. This approach has been tested in three key phytolith genera for the study of agricultural origins in Near East archaeology: Avena, Hordeum and Triticum. Also, this classification has been validated at species-level using Triticum boeoticum and dicoccoides images. Due to the diversity of microscopes, cameras and chemical treatments that can influence images of phytolith slides, three types of data augmentation techniques have been implemented: rotation of the images at 45-degree angles, random colour and brightness jittering, and random blur/sharpen. The implemented workflow has resulted in an overall accuracy of 93.68% for phytolith genera, improving previous attempts. The algorithm has also demonstrated its potential to automatize the classification of phytoliths species with an overall accuracy of 100%. The open code and platforms employed to develop the algorithm assure the method's accessibility, reproducibility and reusability.

Abstract: Despite recent advances in 3D pose estimation of human hands, especially thanks to the advent of CNNs and depth cameras, this task is still far from being solved. This is mainly due to the highly non-linear dynamics of fingers, which make hand model training a challenging task. In this paper, we exploit a novel hierarchical tree-like structured CNN, in which branches are trained to become specialized in predefined subsets of hand joints, called local poses. We further fuse local pose features, extracted from hierarchical CNN branches, to learn higher order dependencies among joints in the final pose by end-to-end training. Lastly, the loss function used is also defined to incorporate appearance and physical constraints about doable hand motion and deformation. Finally, we introduce a non-rigid data augmentation approach to increase the amount of training depth data. Experimental results suggest that feeding a tree-shaped CNN, specialized in local poses, into a fusion network for modeling joints correlations and dependencies, helps to increase the precision of final estimations, outperforming state-of-the-art results on NYU and SyntheticHand datasets.