Semantic segmentation is key in autonomous driving. Using deep visual learning architectures is not trivial in this context, because of the challenges in creating suitable large scale annotated datasets. This issue has been traditionally circumvented through the use of synthetic datasets, that have become a popular resource in this field. They have been released with the need to develop semantic segmentation algorithms able to close the visual domain shift between the training and test data. Although exacerbated by the use of artificial data, the problem is extremely relevant in this field even when training on real data. Indeed, weather conditions, viewpoint changes and variations in the city appearances can vary considerably from car to car, and even at test time for a single, specific vehicle. How to deal with domain adaptation in semantic segmentation, and how to leverage effectively several different data distributions (source domains) are important research questions in this field. To support work in this direction, this paper contributes a new large scale, synthetic dataset for semantic segmentation with more than 100 different source visual domains. The dataset has been created to explicitly address the challenges of domain shift between training and test data in various weather and view point conditions, in seven different city types. Extensive benchmark experiments assess the dataset, showcasing open challenges for the current state of the art. The dataset will be available at: https://idda-dataset.github.io/home/ .
Wearable cameras are becoming more and more popular in several applications, increasing the interest of the research community in developing approaches for recognizing actions from a first-person point of view. An open challenge is how to cope with the limited amount of motion information available about the action itself, as opposed to the more investigated third-person action recognition scenario. When focusing on manipulation tasks, videos tend to record only parts of the movement, making crucial the understanding of the objects being manipulated and of their context. Previous works addressed this issue with two-stream architectures, one dedicated to modeling the appearance of objects involved in the action, another dedicated to extracting motion features from optical flow. In this paper, we argue that features from these two information channels should be learned jointly to capture the spatio-temporal correlations between the two in a better way. To this end, we propose a single stream architecture able to do so, thanks to the addition of a self-supervised block that uses a pretext motion segmentation task to intertwine motion and appearance knowledge. Experiments on several publicly available databases show the power of our approach.
Despite their effectiveness in a wide range of tasks, deep architectures suffer from some important limitations. In particular, they are vulnerable to catastrophic forgetting, i.e. they perform poorly when they are required to update their model as new classes are available but the original training set is not retained. This paper addresses this problem in the context of semantic segmentation. Current strategies fail on this task because they do not consider a peculiar aspect of semantic segmentation: since each training step provides annotation only for a subset of all possible classes, pixels of the background class (i.e. pixels that do not belong to any other classes) exhibit a semantic distribution shift. In this work we revisit classical incremental learning methods, proposing a new distillation-based framework which explicitly accounts for this shift. Furthermore, we introduce a novel strategy to initialize classifier's parameters, thus preventing biased predictions toward the background class. We demonstrate the effectiveness of our approach with an extensive evaluation on the Pascal-VOC 2012 and ADE20K datasets, significantly outperforming state of the art incremental learning methods.
Although deep networks have significantly increased the performance of visual recognition methods, it is still challenging to achieve the robustness across visual domains that is necessary for real-world applications. To tackle this issue, research on domain adaptation and generalization has flourished over the last decade. An important aspect to consider when assessing the work done in the literature so far is the amount of data annotation necessary for training each approach, both at the source and target level. In this paper we argue that the data annotation overload should be minimal, as it is costly. Hence, we propose to use self-supervised learning to achieve domain generalization and adaptation. We consider learning regularities from non annotated data as an auxiliary task, and cast the problem within an Auxiliary Learning principled framework. Moreover, we suggest to further exploit the ability to learn about visual domains from non annotated images by learning from target data while testing, as data are presented to the algorithm one sample at a time. Results on three different scenarios confirm the value of our approach.
While today's robots are able to perform sophisticated tasks, they can only act on objects they have been trained to recognize. This is a severe limitation: any robot will inevitably see new objects in unconstrained settings, and thus will always have visual knowledge gaps. However, standard visual modules are usually built on a limited set of classes and are based on the strong prior that an object must belong to one of those classes. Identifying whether an instance does not belong to the set of known categories (i.e. open set recognition), only partially tackles this problem, as a truly autonomous agent should be able not only to detect what it does not know, but also to extend dynamically its knowledge about the world. We contribute to this challenge with a deep learning architecture that can dynamically update its known classes in an end-to-end fashion. The proposed deep network, based on a deep extension of a non-parametric model, detects whether a perceived object belongs to the set of categories known by the system and learns it without the need to retrain the whole system from scratch. Annotated images about the new category can be provided by an 'oracle' (i.e. human supervision), or by autonomous mining of the Web. Experiments on two different databases and on a robot platform demonstrate the promise of our approach.
Human adaptability relies crucially on the ability to learn and merge knowledge both from supervised and unsupervised learning: the parents point out few important concepts, but then the children fill in the gaps on their own. This is particularly effective, because supervised learning can never be exhaustive and thus learning autonomously allows to discover invariances and regularities that help to generalize. In this paper we propose to apply a similar approach to the task of object recognition across domains: our model learns the semantic labels in a supervised fashion, and broadens its understanding of the data by learning from self-supervised signals how to solve a jigsaw puzzle on the same images. This secondary task helps the network to learn the concepts of spatial correlation while acting as a regularizer for the classification task. Multiple experiments on the PACS, VLCS, Office-Home and digits datasets confirm our intuition and show that this simple method outperforms previous domain generalization and adaptation solutions. An ablation study further illustrates the inner workings of our approach.
Deep networks have brought significant advances in robot perception, enabling to improve the capabilities of robots in several visual tasks, ranging from object detection and recognition to pose estimation, semantic scene segmentation and many others. Still, most approaches typically address visual tasks in isolation, resulting in overspecialized models which achieve strong performances in specific applications but work poorly in other (often related) tasks. This is clearly sub-optimal for a robot which is often required to perform simultaneously multiple visual recognition tasks in order to properly act and interact with the environment. This problem is exacerbated by the limited computational and memory resources typically available onboard to a robotic platform. The problem of learning flexible models which can handle multiple tasks in a lightweight manner has recently gained attention in the computer vision community and benchmarks supporting this research have been proposed. In this work we study this problem in the robot vision context, proposing a new benchmark, the RGB-D Triathlon, and evaluating state of the art algorithms in this novel challenging scenario. We also define a new evaluation protocol, better suited to the robot vision setting. Results shed light on the strengths and weaknesses of existing approaches and on open issues, suggesting directions for future research.
The ability to categorize is a cornerstone of visual intelligence, and a key functionality for artificial, autonomous visual machines. This problem will never be solved without algorithms able to adapt and generalize across visual domains. Within the context of domain adaptation and generalization, this paper focuses on the predictive domain adaptation scenario, namely the case where no target data are available and the system has to learn to generalize from annotated source images plus unlabeled samples with associated metadata from auxiliary domains. Our contributionis the first deep architecture that tackles predictive domainadaptation, able to leverage over the information broughtby the auxiliary domains through a graph. Moreover, we present a simple yet effective strategy that allows us to take advantage of the incoming target data at test time, in a continuous domain adaptation scenario. Experiments on three benchmark databases support the value of our approach.
Visual recognition systems are meant to work in the real world. For this to happen, they must work robustly in any visual domain, and not only on the data used during training. Within this context, a very realistic scenario deals with domain generalization, i.e. the ability to build visual recognition algorithms able to work robustly in several visual domains, without having access to any information about target data statistic. This paper contributes to this research thread, proposing a deep architecture that maintains separated the information about the available source domains data while at the same time leveraging over generic perceptual information. We achieve this by introducing domain-specific aggregation modules that through an aggregation layer strategy are able to merge generic and specific information in an effective manner. Experiments on two different benchmark databases show the power of our approach, reaching the new state of the art in domain generalization.