Abstract: Food image recognition is a complex computer vision task, because of the large number of fine-grained food classes. Fine-grained recognition tasks focus on learning subtle discriminative details to distinguish similar classes. In this paper, we introduce a new method to improve the classification of classes that are more difficult to discriminate based on Multi-Subsets learning. Using a pre-trained network, we organize classes in multiple subsets using a clustering technique. Later, we embed these subsets in a multi-head model structure. This structure has three distinguishable parts. First, we use several shared blocks to learn the generalized representation of the data. Second, we use multiple specialized blocks focusing on specific subsets that are difficult to distinguish. Lastly, we use a fully connected layer to weight the different subsets in an end-to-end manner by combining the neuron outputs. We validated our proposed method using two recent state-of-the-art vision transformers on three public food recognition datasets. Our method was successful in learning the confused classes better and we outperformed the state-of-the-art on the three datasets.

Abstract: Until recently, most computer vision models remained illiterate, largely ignoring the semantically rich and explicit information contained in scene text. Recent progress in scene text detection and recognition has recently allowed exploring its role in a diverse set of open computer vision problems, e.g. image classification, image-text retrieval, image captioning, and visual question answering to name a few. The explicit semantics of scene text closely requires specific modeling similar to language. However, scene text is a particular signal that has to be interpreted according to a comprehensive perspective that encapsulates all the visual cues in an image. Incorporating this information is a straightforward task for humans, but if we are unfamiliar with a language or scripture, achieving a complete world understanding is impossible (e.a. visiting a foreign country with a different alphabet). Despite the importance of scene text, modeling it requires considering the several ways in which scene text interacts with an image, processing and fusing an additional modality. In this thesis, we mainly focus
on two tasks, scene text-based fine-grained image classification, and cross-modal retrieval. In both studied tasks we identify existing limitations in current approaches and propose plausible solutions. Concretely, in each chapter: i) We define a compact way to embed scene text that generalizes to unseen words at training time while performing in real-time. ii) We incorporate the previously learned scene text embedding to create an image-level descriptor that overcomes optical character recognition (OCR) errors which is well-suited to the fine-grained image classification task. iii) We design a region-level reasoning network that learns the interaction through semantics among salient visual regions and scene text instances. iv) We employ scene text information in image-text matching and introduce the Scene Text Aware Cross-Modal retrieval StacMR task. We gather a dataset that incorporates scene text and design a model suited for the newly studied modality. v) We identify the drawbacks of current retrieval metrics in cross-modal retrieval. An image captioning metric is proposed as a way of better evaluating semantics in retrieved results. Ample experimentation shows that incorporating such semantics into a model yields better semantic results while
requiring significantly less data to converge.

Abstract: Hand-written Text Recognition techniques with the aim to automatically identify and transcribe hand-written text have been applied to historical sources including ciphers. In this paper, we compare the performance of two machine learning architectures, an unsupervised method based on clustering and a deep learning method with few-shot learning. Both models are tested on seen and unseen data from historical ciphers with different symbol sets consisting of various types of graphic signs. We compare the models and highlight their differences in performance, with their advantages and shortcomings.

Abstract: Radiomics is an emerging technique for the quantification of imaging data that has recently shown great promise for deeper phenotyping of cardiovascular disease. Thus far, the technique has been mostly applied in single-centre studies. However, one of the main difficulties in multi-centre imaging studies is the inherent variability of image characteristics due to centre differences. In this paper, a comprehensive analysis of radiomics variability under several image- and feature-based normalisation techniques was conducted using a multi-centre cardiovascular magnetic resonance dataset. 218 subjects divided into healthy (n = 112) and hypertrophic cardiomyopathy (n = 106, HCM) groups from five different centres were considered. First and second order texture radiomic features were extracted from three regions of interest, namely the left and right ventricular cavities and the left ventricular myocardium. Two methods were used to assess features’ variability. First, feature distributions were compared across centres to obtain a distribution similarity index. Second, two classification tasks were proposed to assess: (1) the amount of centre-related information encoded in normalised features (centre identification) and (2) the generalisation ability for a classification model when trained on these features (healthy versus HCM classification). The results showed that the feature-based harmonisation technique ComBat is able to remove the variability introduced by centre information from radiomic features, at the expense of slightly degrading classification performance. Piecewise linear histogram matching normalisation gave features with greater generalisation ability for classification ( balanced accuracy in between 0.78 ± 0.08 and 0.79 ± 0.09). Models trained with features from images without normalisation showed the worst performance overall ( balanced accuracy in between 0.45 ± 0.28 and 0.60 ± 0.22). In conclusion, centre-related information removal did not imply good generalisation ability for classification.

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: This paper details the results and main findings of the second iteration of the Multi-modal Aerial View Object Classification (MAVOC) challenge. The primary goal of both MAVOC challenges is to inspire research into methods for building recognition models that utilize both synthetic aperture radar (SAR) and electro-optical (EO) imagery. Teams are encouraged to develop multi-modal approaches that incorporate complementary information from both domains. While the 2021 challenge showed a proof of concept that both modalities could be used together, the 2022 challenge focuses on the detailed multi-modal methods. The 2022 challenge uses the same UNIfied Coincident Optical and Radar for recognitioN (UNICORN) dataset and competition format that was used in 2021. Specifically, the challenge focuses on two tasks, (1) SAR classification and (2) SAR + EO classification. The bulk of this document is dedicated to discussing the top performing methods and describing their performance on our blind test set. Notably, all of the top ten teams outperform a Resnet-18 baseline. For SAR classification, the top team showed a 129% improvement over baseline and an 8% average improvement from the 2021 winner. The top team for SAR + EO classification shows a 165% improvement with a 32% average improvement over 2021.

Abstract: The sewerage infrastructure is one of the most important and expensive infrastructures in modern society. In order to efficiently manage the sewerage infrastructure, automated sewer inspection has to be utilized. However, while sewer
defect classification has been investigated for decades, little attention has been given to classifying sewer pipe properties such as water level, pipe material, and pipe shape, which are needed to evaluate the level of sewer pipe deterioration.
In this work we classify sewer pipe defects and properties concurrently and present a novel decoder-focused multi-task classification architecture Cross-Task Graph Neural Network (CT-GNN), which refines the disjointed per-task predictions using cross-task information. The CT-GNN architecture extends the traditional disjointed task-heads decoder, by utilizing a cross-task graph and unique class node embeddings. The cross-task graph can either be determined a priori based on the conditional probability between the task classes or determined dynamically using self-attention.
CT-GNN can be added to any backbone and trained end-toend at a small increase in the parameter count. We achieve state-of-the-art performance on all four classification tasks in the Sewer-ML dataset, improving defect classification and
water level classification by 5.3 and 8.0 percentage points, respectively. We also outperform the single task methods as well as other multi-task classification approaches while introducing 50 times fewer parameters than previous modelfocused approaches.