Abstract: 36.307
Many everyday behavioural uses of color vision depend on color naming ability, which is neither measured nor predicted by most standardized tests of color vision, for either normal or anomalous color vision. Here we demonstrate a new method to quantify color naming ability by deriving a compact computational description of individual 3D color spaces. Methods: Individual observers underwent standardized color vision diagnostic tests (including anomaloscope testing) and a series of custom-made color naming tasks using 500 distinct color samples, either CRT stimuli (“light”-based) or Munsell chips (“surface”-based), with both forced- and free-choice color naming paradigms. For each subject, we defined his/her color solid as the set of 3D convex hulls computed for each basic color category from the relevant collection of categorised points in perceptually uniform CIELAB space. From the parameters of the convex hulls, we derived several indices to characterise the 3D structure of the color solid and its inter-individual variations. Using a reference group of 25 normal trichromats (NT), we defined the degree of normality for the shape, location and overlap of each color region, and the extent of “light”-“surface” agreement. Results: Certain features of color perception emerge from analysis of the average NT color solid, e.g.: (1) the white category is slightly shifted towards blue; and (2) the variability in category border location across NT subjects is asymmetric across color space, with least variability in the blue/green region. Comparisons between individual and average NT indices reveal specific naming “deficits”, e.g.: (1) Category volumes for white, green, brown and grey are expanded for anomalous trichromats and dichromats; and (2) the focal structure of color space is disrupted more in protanopia than other forms of anomalous color vision. The indices both capture the structure of subjective color spaces and allow us to quantify inter-individual differences in color naming ability.

Abstract: 3D geometric information is essential for on-board perception in autonomous driving and driver assistance. Autonomous vehicles (AVs) are equipped with calibrated sensor suites. As part of these suites, we can find LiDARs, which are expensive active sensors in charge of providing the 3D geometric information. Depending on the operational conditions for the AV, calibrated stereo rigs may be also sufficient for obtaining 3D geometric information, being these rigs less expensive and easier to install than LiDARs. However, ensuring a proper maintenance and calibration of these types of sensors is not trivial. Accordingly, there is an increasing interest on performing monocular depth estimation (MDE) to obtain 3D geometric information on-board. MDE is very appealing since it allows for appearance and depth being on direct pixelwise correspondence without further calibration. Moreover, a set of single cameras with MDE capabilities would still be a cheap solution for on-board perception, relatively easy to integrate and maintain in an AV.
Best MDE models are based on Convolutional Neural Networks (CNNs) trained in a supervised manner, i.e., assuming pixelwise ground truth (GT). Accordingly, the overall goal of this PhD is to study methods for improving CNN-based MDE accuracy under different training settings. More specifically, this PhD addresses different research questions that are described below. When we started to work in this PhD, state-of-theart methods for MDE were already based on CNNs. In fact, a promising line of work consisted in using image-based semantic supervision (i.e., pixel-level class labels) while training CNNs for MDE using LiDAR-based supervision (i.e., depth). It was common practice to assume that the same raw training data are complemented by both types of supervision, i.e., with depth and semantic labels. However, in practice, it was more common to find heterogeneous datasets with either only depth supervision or only semantic supervision. Therefore, our first work was to research if we could train CNNs for MDE by leveraging depth and semantic information from heterogeneous datasets. We show that this is indeed possible, and we surpassed the state-of-the-art results on MDE at the time we did this research. To achieve our results, we proposed a particular CNN architecture and a new training protocol.
After this research, it was clear that the upper-bound setting to train CNN-based MDE models consists in using LiDAR data as supervision. However, it would be cheaper and more scalable if we would be able to train such models from monocular sequences. Obviously, this is far more challenging, but worth to research. Training MDE models using monocular sequences is possible by relying on structure-from-motion (SfM) principles to generate self-supervision. Nevertheless, problems of camouflaged objects, visibility changes, static-camera intervals, textureless areas, and scale ambiguity, diminish the usefulness of such self-supervision. To alleviate these problems, we perform MDE by virtual-world supervision and real-world SfM self-supervision. We call our proposalMonoDEVSNet. We compensate the SfM self-supervision limitations by leveraging
virtual-world images with accurate semantic and depth supervision, as well as addressing the virtual-to-real domain gap. MonoDEVSNet outperformed previous MDE CNNs trained on monocular and even stereo sequences. We have publicly released MonoDEVSNet at <https://github.com/HMRC-AEL/MonoDEVSNet>.
Finally, since MDE is performed to produce 3D information for being used in downstream tasks related to on-board perception. We also address the question of whether the standard metrics for MDE assessment are a good indicator for future MDE-based driving-related perception tasks. By using 3D object detection on point clouds as proxy of on-board perception, we conclude that, indeed, MDE evaluation metrics give rise to a ranking of methods which reflects relatively well the 3D object detection results we may expect.

Abstract: 5Y Impact Factor: 1.14 / 3rd (Computer Science, Artificial Intelligence)
We introduce a model of attentional eye guidance based on the rationale that the deployment of gaze is to be considered in the context of a general action-perception loop relying on two strictly intertwined processes: sensory processing, depending on current gaze position, identifies sources of information that are most valuable under the given task; motor processing links such information with the oculomotor act by sampling the next gaze position and thus performing the gaze shift. In such a framework, the choice of where to look next is task-dependent and oriented to classes of objects embedded within pictures of complex scenes. The dependence on task is taken into account by exploiting the value and the payoff of gazing at certain image patches or proto-objects that provide a sparse representation of the scene objects. The different levels of the action-perception loop are represented in probabilistic form and eventually give rise to a stochastic process that generates the gaze sequence. This way the model also accounts for statistical properties of gaze shifts such as individual scan path variability. Results of the simulations are compared either with experimental data derived from publicly available datasets and from our own experiments.

Abstract: : Edge-based color constancy methods make use of image derivatives to estimate the illuminant. However, different edge types exist in real-world images such as material, shadow and highlight edges. These different edge types may have a distinctive influence on the performance of the illuminant estimation. Therefore, in this paper, an extensive analysis is provided of different edge types on the performance of edge-based color constancy methods. First, an edge-based taxonomy is presented classifying edge types based on their photometric properties (e.g. material, shadow-geometry and highlights). Then, a performance evaluation of edge-based color constancy is provided using these different edge types. From this performance evaluation it is derived that specular and shadow edge types are more valuable than material edges for the estimation of the illuminant. To this end, the (iterative) weighted Grey-Edge algorithm is proposed in which these edge types are more emphasized for the estimation of the illuminant. Images that are recorded under controlled circumstances demonstrate that the proposed iterative weighted Grey-Edge algorithm based on highlights reduces the median angular error with approximately $25\%$. In an uncontrolled environment, improvements in angular error up to $11\%$ are obtained with respect to regular edge-based color constancy.

Abstract: A benchmark of saliency models performance with a synthetic image dataset is provided. Model performance is evaluated through saliency metrics as well as the influence of model inspiration and consistency with human psychophysics. SID4VAM is composed of 230 synthetic images, with known salient regions. Images were generated with 15 distinct types of low-level features (e.g. orientation, brightness, color, size...) with a target-distractor pop-out type of synthetic patterns. We have used Free-Viewing and Visual Search task instructions and 7 feature contrasts for each feature category. Our study reveals that state-of-the-art Deep Learning saliency models do not perform well with synthetic pattern images, instead, models with Spectral/Fourier inspiration outperform others in saliency metrics and are more consistent with human psychophysical experimentation. This study proposes a new way to evaluate saliency models in the forthcoming literature, accounting for synthetic images with uniquely low-level feature contexts, distinct from previous eye tracking image datasets.

Abstract: A common technique to avoid overfitting when training deep neural networks (DNN) is to monitor the performance in a dedicated validation data partition and to stop
training as soon as it saturates. This only focuses on what the model does, while completely ignoring what happens inside it.
In this work, we open the “black-box” of DNN in order to perform early stopping. We propose to use a novel theoretical framework that analyses meso-scale patterns in the topology of the functional graph of a network while it trains. Based on it,
we decide when it transitions from learning towards overfitting in a more explainable way. We exemplify the benefits of this approach on a state-of-the art custom DNN that jointly learns local representations and label structure employing an ensemble of dedicated subnetworks. We show that it is practically equivalent in performance to early stopping with patience, the standard early stopping algorithm in the literature. This proves beneficial for AU recognition performance and provides new insights into how learning of AUs occurs in DNNs.

Abstract: A common way to model multiclass classification problems is to design a set of binary classifiers and to combine them. Error-correcting output codes (ECOC) represent a successful framework to deal with these type of problems. Recent works in the ECOC framework showed significant performance improvements by means of new problem-dependent designs based on the ternary ECOC framework. The ternary framework contains a larger set of binary problems because of the use of a ldquodo not carerdquo symbol that allows us to ignore some classes by a given classifier. However, there are no proper studies that analyze the effect of the new symbol at the decoding step. In this paper, we present a taxonomy that embeds all binary and ternary ECOC decoding strategies into four groups. We show that the zero symbol introduces two kinds of biases that require redefinition of the decoding design. A new type of decoding measure is proposed, and two novel decoding strategies are defined. We evaluate the state-of-the-art coding and decoding strategies over a set of UCI machine learning repository data sets and into a real traffic sign categorization problem. The experimental results show that, following the new decoding strategies, the performance of the ECOC design is significantly improved.