Abstract: Computer-aided diagnosis (CAD) is a tool with great potential to help endoscopists in the tasks of detecting and histologically classifying colorectal polyps. In recent years, different technologies have been described and their potential utility has been increasingly evidenced, which has generated great expectations among scientific societies. However, most of these works are retrospective and use images of different quality and characteristics which are analysed off line. This review aims to familiarise gastroenterologists with computational methods and the particularities of endoscopic imaging, which have an impact on image processing analysis. Finally, the publicly available image databases, needed to compare and confirm the results obtained with different methods, are presented.

Abstract: Chapter 9

Abstract: The curse of data labeling is a costly bottleneck in supervised deep learning, where large amounts of labeled data are needed to train intelligent systems. In onboard perception for autonomous driving, this cost corresponds to the labeling of raw data from sensors such as cameras, LiDARs, RADARs, etc. Therefore, synthetic data with automatically generated ground truth (labels) has aroused as a reliable alternative for training onboard perception models.
However, synthetic data commonly suffers from synth-to-real domain shift, i.e., models trained on the synthetic domain do not show their achievable accuracy when performing in the real world. This shift needs to be addressed by techniques falling in the realm of domain adaptation (DA).
The semi-supervised learning (SSL) paradigm can be followed to address DA. In this case, a model is trained using source data with labels (here synthetic) and leverages minimal knowledge from target data (here the real world) to generate pseudo-labels. These pseudo-labels help the training process to reduce the gap between the source and the target domains. In general, we can assume accessing both, pseudo-labels and a few amounts of human-provided labels for the target-domain data. However, the most interesting and challenging setting consists in assuming that we do not have human-provided labels at all. This setting is known as unsupervised domain adaptation (UDA). This PhD focuses on applying SSL to the UDA setting, for onboard visual tasks related to autonomous driving. We start by addressing the synth-to-real UDA problem on onboard vision-based object detection (pedestrians and cars), a critical task for autonomous driving and driving assistance. In particular, we propose to apply an SSL technique known as co-training, which we adapt to work with deep models that process a multi-modal input. The multi-modality consists of the visual appearance of the images (RGB) and their monocular depth estimation. The synthetic data we use as the source domain contains both, object bounding boxes and depth information. This prior knowledge is the
starting point for the co-training technique, which iteratively labels unlabeled real-world data and uses such pseudolabels (here bounding boxes with an assigned object class) to progressively improve the labeling results. Along this
process, two models collaborate to automatically label the images, in a way that one model compensates for the errors of the other, so avoiding error drift. While this automatic labeling process is done offline, the resulting pseudolabels can be used to train object detection models that must perform in real-time onboard a vehicle. We show that multi-modal co-training improves the labeling results compared to single-modal co-training, remaining competitive compared to human labeling.
Given the success of co-training in the context of object detection, we have also adapted this technique to a more crucial and challenging visual task, namely, onboard semantic segmentation. In fact, providing labels for a single image
can take from 30 to 90 minutes for a human labeler, depending on the content of the image. Thus, developing automatic labeling techniques for this visual task is of great interest to the automotive industry. In particular, the new co-training framework addresses synth-to-real UDA by an initial stage of self-training. Intermediate models arising from this stage are used to start the co-training procedure, for which we have elaborated an accurate collaboration policy between the two models performing the automatic labeling. Moreover, our co-training seamlessly leverages datasets from different synthetic domains. In addition, the co-training procedure is agnostic to the loss function used to train the semantic segmentation models which perform the automatic labeling. We achieve state-of-the-art results on publicly available benchmark datasets, again, remaining competitive compared to human labeling.
Finally, on the ground of our previous experience, we have designed and implemented a new SSL technique for UDA in the context of visual semantic segmentation. In this case, we mimic the labeling methodology followed by human labelers. In particular, rather than labeling full images at a time, categories of semantic classes are defined and only those are labeled in a labeling pass. In fact, different human labelers can become specialists in labeling different categories. Afterward, these per-category-labeled layers are combined to provide fully labeled images. Our technique is inspired by this methodology since we perform synth-to-real UDA per category, using the self-training stage previously developed as part of our co-training framework. The pseudo-labels obtained for each category are finally
fused to obtain fully automatically labeled images. In this context, we have also contributed to the development of a new photo-realistic synthetic dataset based on path-tracing rendering. Our new SSL technique seamlessly leverages publicly available synthetic datasets as well as this new one to obtain state-of-the-art results on synth-to-real UDA for semantic segmentation. We show that the new dataset allows us to reach better labeling accuracy than previously existing datasets, at the same time that it complements well them when combined. Moreover, we also show that the new human-inspired SSL technique outperforms co-training.

Abstract: Robust and dense computation of depth information from stereo-camera systems is a computationally demanding requirement for real-time autonomous driving. Semi-Global Matching (SGM) [1] approximates heavy-computation global algorithms results but with lower computational complexity, therefore it is a good candidate for a real-time implementation. SGM minimizes energy along several 1D paths across the image. The aim of this work is to provide a real-time system producing reliable results on energy-efficient hardware. Our design runs on a NVIDIA Titan X GPU at 104.62 FPS and on a NVIDIA Drive PX at 6.7 FPS, promising for real-time platforms

Abstract: In pattern recognition [8, 14], a key issue to be addressed when designing a system is how to represent input patterns. Feature vectors is a common option. That is, a set of numerical features describing relevant properties of the pattern are computed and arranged in a vector form. The main advantages of this kind of representation are computational simplicity and a well sound mathematical foundation. Thus, a large number of operations are available to work with vectors and a large repository of algorithms for pattern analysis and classification exist. However, the simple structure of feature vectors might not be the best option for complex patterns where nonnumerical features or relations between different parts of the pattern become relevant.

Abstract: This paper deals with the recognition of symbols and structural textures in architectural plans using string matching techniques. A plan is represented by an attributed graph whose nodes represent characteristic points and whose edges represent segments. Symbols and textures can be seen as a set of regions, i.e. closed loops in the graph, with a particular arrangement. The search for a symbol involves a graph matching between the regions of a model graph and the regions of the graph representing the document. Discriminating a texture means a clustering of neighbouring regions of this graph. Both procedures involve a similarity measure between graph regions. A string codification is used to represent the sequence of outlining edges of a region. Thus, the similarity between two regions is defined in terms of the string edit distance between their boundary strings. The use of string matching allows the recognition method to work also under presence of distortion.