Panoptic segmentation is the recently introduced task that tackles semantic segmentation and instance segmentation jointly [18]. In this paper, we present an extension of SemanticKITTI [1], a large-scale dataset providing dense point-wise semantic labels for all sequences of the KITTI Odometry Benchmark [10]. This extension enables training and evaluation of LiDAR-based panoptic segmentation. We provide the data and discuss the processing steps needed to enrich a given semantic annotation with temporally consistent instance information, i.e., instance information that supplements the semantic labels and identifies the same instance over sequences of LiDAR point clouds. Additionally, we present two strong baselines that combine state-of-the-art LiDAR-based semantic segmentation approaches with a state-of-the-art detector enriching the segmentation with instance information and that allow other researchers to compare their approaches against. We believe that our extension of SemanticKITTI with strong baselines enables the creation of novel algorithms for LiDAR-based panoptic segmentation as much as it has for the original semantic segmentation and semantic scene completion tasks. Data, code, and an online evaluation service using a hidden test set are publicly available at http://semantic-kitti.org.

Knowing the distance to nearby objects is crucial for autonomous cars to navigate safely in everyday traffic. In this paper, we investigate monocular depth estimation, which advanced substantially within the last years and is providing increasingly more accurate results while only requiring a single camera image as input. In line with recent work, we use an encoder-decoder structure with so-called packing layers to estimate depth values in a self-supervised fashion. We propose integrating a joint pre-training of semantic segmentation plus depth estimation on a dataset providing semantic labels. By using a separate semantic decoder that is only needed for pre-training, we can keep the network comparatively small. Our extensive experimental evaluation shows that the addition of such pre-training improves the depth estimation performance substantially. Finally, we show that we achieve competitive performance on the KITTI dataset despite using a much smaller and more efficient network.

Annotated datasets are essential for supervised learning. However, annotating large datasets is a tedious and time-intensive task. This paper addresses active learning in the context of semantic segmentation with the goal of reducing the human labeling effort. Our application is agricultural robotics and we focus on the task of distinguishing between crop and weed plants from image data. A key challenge in this application is the transfer of an existing semantic segmentation CNN to a new field, in which growth stage, weeds, soil, and weather conditions differ. We propose a novel approach that, given a trained model on one field together with rough foreground segmentation, refines the network on a substantially different field providing an effective method of selecting samples to annotate for supporting the transfer. We evaluated our approach on two challenging datasets from the agricultural robotics domain and show that we achieve a higher accuracy with a smaller number of samples compared to random sampling as well as entropy based sampling, which consequently reduces the required human labeling effort.

Inferring semantic information towards an understanding of the surrounding environment is crucial for autonomous vehicles to drive safely. Deep learning-based segmentation methods can infer semantic information directly from laser range data, even in the absence of other sensor modalities such as cameras. In this paper, we address improving the generalization capabilities of such deep learning models to range data that was captured using a different sensor and in situations where no labeled data is available for the new sensor setup. Our approach assists the domain transfer of a LiDAR-only semantic segmentation model to a different sensor and environment exploiting existing geometric mapping systems. To this end, we fuse sequential scans in the source dataset into a dense mesh and render semi-synthetic scans that match those of the target sensor setup. Unlike simulation, this approach provides a real-to-real transfer of geometric information and delivers additionally more accurate remission information. We implemented and thoroughly tested our approach by transferring semantic scans between two different real-world datasets with different sensor setups. Our experiments show that we can improve the segmentation performance substantially with zero manual re-labeling. This approach solves the number one feature request since we released our semantic segmentation library LiDAR-bonnetal [18].

Truly autonomous driving without the need for human intervention can only be attained when self-driving cars fully understand their surroundings. Most of these vehicles rely on a suite of active and passive sensors. LiDAR sensors are a cornerstone in most of these hardware stacks, and leveraging them as a complement to other passive sensors such as RGB cameras is an enticing goal. Understanding the semantic class of each point in a LiDAR sweep is important, as well as knowing to which instance of that class it belongs to. To this end, we present a novel, single-stage, and real-time capable panoptic segmentation approach using a shared encoder with a semantic and instance decoder. We leverage the geometric information of the LiDAR scan to perform a novel, distance- aware tri-linear upsampling, which allows our approach to use larger output strides than using transpose convolutions leading to substantial savings in computation time. Our experimental evaluation and ablation studies for each module show that combining our geometric and semantic embeddings with our learned, variable instance thresholds, a category-specific loss, and the novel trilinear upsampling module leads to higher panoptic quality. We will release the code of our approach in our LiDAR processing library LiDAR-Bonnetal [27].

Semantic scene understanding is important for various applications. In particular, self-driving cars need a fine-grained understanding of the surfaces and objects in their vicinity. Light detection and ranging (LiDAR) provides precise geometric information about the environment and is thus a part of the sensor suites of almost all self-driving cars. Despite the relevance of semantic scene understanding for this application, there is a lack of a large dataset for this task which is based on an automotive LiDAR. In this paper, we introduce a large dataset to propel research on laser-based semantic segmentation. We annotated all sequences of the KITTI Vision Odometry Benchmark and provide dense point-wise annotations for the complete 360-degree field-of-view of the employed automotive LiDAR. We propose three benchmark tasks based on this dataset: (i) semantic segmentation of point clouds using a single scan, (ii) semantic segmentation using multiple past scans, and (iii) semantic scene completion, which requires to anticipate the semantic scene in the future. We provide baseline experiments and show that there is a need for more sophisticated models to efficiently tackle these tasks. Our dataset opens the door for the development of more advanced methods, but also provides plentiful data to investigate new research directions.

Semantic scene understanding is important for autonomous robots that aim to navigate dynamic environments, manipulate objects, or interact with humans in a natural way. In this paper, we address the problem of jointly performing semantic segmentation as well as instance segmentation in an online fashion, so that autonomous robots can use this information on-the-go and without sacrificing accuracy. We achieve this by exploiting a local connectivity prior of objects in the real world and a multi-task convolutional neural network architecture. The network identifies the individual object instances and their classes without region proposals or pre-segmentation of the images into individual classes. We implemented and thoroughly evaluated our approach, and our experiments suggest that our method can be used to accurately segment instance masks of objects and identify their class in an online fashion.

The ability to interpret a scene is an important capability for a robot that is supposed to interact with its environment. The knowledge of what is in front of the robot is, for example, relevant for navigation, manipulation, or planning. Semantic segmentation labels each pixel of an image with a class label and thus provides a detailed semantic annotation of the surroundings to the robot. Convolutional neural networks (CNNs) are popular methods for addressing this type of problem. The available software for training and the integration of CNNs for real robots, however, is quite fragmented and often difficult to use for non-experts, despite the availability of several high-quality open-source frameworks for neural network implementation and training. In this paper, we propose a tool called Bonnet, which addresses this fragmentation problem by building a higher abstraction that is specific for the semantic segmentation task. It provides a modular approach to simplify the training of a semantic segmentation CNN independently of the used dataset and the intended task. Furthermore, we also address the deployment on a real robotic platform. Thus, we do not propose a new CNN approach in this paper. Instead, we provide a stable and easy-to-use tool to make this technology more approachable in the context of autonomous systems. In this sense, we aim at closing a gap between computer vision research and its use in robotics research. We provide an open-source codebase for training and deployment. The training interface is implemented in Python using TensorFlow and the deployment interface provides C++ library that can be easily integrated in an existing robotics codebase, a ROS node, and two standalone applications for label prediction in images and videos.

Reliable and accurate localization and mapping are key components of most autonomous systems. Besides geometric information about the mapped environment, the semantics plays an important role to enable intelligent navigation behaviors. In most realistic environments, this task is particularly complicated due to dynamics caused by moving objects, which can corrupt the mapping step or derail localization. In this paper, we propose an extension of a recently published surfel-based mapping approach exploiting three-dimensional laser range scans by integrating semantic information to facilitate the mapping process. The semantic information is efficiently extracted by a fully convolutional neural network and rendered on a spherical projection of the laser range data. This computed semantic segmentation results in point-wise labels for the whole scan, allowing us to build a semantically-enriched map with labeled surfels. This semantic map enables us to reliably filter moving objects, but also improve the projective scan matching via semantic constraints. Our experimental evaluation on challenging highways sequences from KITTI dataset with very few static structures and a large amount of moving cars shows the advantage of our semantic SLAM approach in comparison to a purely geometric, state-of-the-art approach.

Perception in autonomous vehicles is often carried out through a suite of different sensing modalities. Given the massive amount of openly available labeled RGB data and the advent of high-quality deep learning algorithms for image-based recognition, high-level semantic perception tasks are pre-dominantly solved using high-resolution cameras. As a result of that, other sensor modalities potentially useful for this task are often ignored. In this paper, we push the state of the art in LiDAR-only semantic segmentation forward in order to provide another independent source of semantic information to the vehicle. Our approach can accurately perform full semantic segmentation of LiDAR point clouds at sensor frame rate. We exploit range images as an intermediate representation in combination with a Convolutional Neural Network (CNN) exploiting the rotating LiDAR sensor model. To obtain accurate results, we propose a novel post-processing algorithm that deals with problems arising from this intermediate representation such as discretization errors and blurry CNN outputs. We implemented and thoroughly evaluated our approach including several comparisons to the state of the art. Our experiments show that our approach outperforms state-of-the-art approaches, while still running online on a single embedded GPU. The code can be accessed at https://github.com/PRBonn/lidar-bonnetal.

Precision farming robots, which target to reduce the amount of herbicides that need to be brought out in the fields, must have the ability to identify crops and weeds in real time to trigger weeding actions. In this paper, we address the problem of CNN-based semantic segmentation of crop fields separating sugar beet plants, weeds, and background solely based on RGB data. We propose a CNN that exploits existing vegetation indexes and provides a classification in real time. Furthermore, it can be effectively re-trained to so far unseen fields with a comparably small amount of training data. We implemented and thoroughly evaluated our system on a real agricultural robot operating in different fields in Germany and Switzerland. The results show that our system generalizes well, can operate at around 20 Hz, and is suitable for online operation in the fields.

Global climate change plays an essential role in our daily life. Mesoscale ocean eddies have a significant impact on global warming, since they affect the ocean dynamics, the energy as well as the mass transports of ocean circulation. From satellite altimetry we can derive high-resolution, global maps containing ocean signals with dominating coherent eddy structures. The aim of this study is the development and evaluation of a deep-learning based approach for the analysis of eddies. In detail, we develop an eddy identification and tracking framework with two different approaches that are mainly based on feature learning with convolutional neural networks. Furthermore, state-of-the-art image processing tools and object tracking methods are used to support the eddy tracking. In contrast to previous methods, our framework is able to learn a representation of the data in which eddies can be detected and tracked in more objective and robust way. We show the detection and tracking results on sea level anomalies (SLA) data from the area of Australia and the East Australia current, and compare our two eddy detection and tracking approaches to identify the most robust and objective method.

Applying agrochemicals is the default procedure for conventional weed control in crop production, but has negative impacts on the environment. Robots have the potential to treat every plant in the field individually and thus can reduce the required use of such chemicals. To achieve that, robots need the ability to identify crops and weeds in the field and must additionally select effective treatments. While certain types of weed can be treated mechanically, other types need to be treated by (selective) spraying. In this paper, we present an approach that provides the necessary information for effective plant-specific treatment. It outputs the stem location for weeds, which allows for mechanical treatments, and the covered area of the weed for selective spraying. Our approach uses an end-to-end trainable fully convolutional network that simultaneously estimates stem positions as well as the covered area of crops and weeds. It jointly learns the class-wise stem detection and the pixel-wise semantic segmentation. Experimental evaluations on different real-world datasets show that our approach is able to reliably solve this problem. Compared to state-of-the-art approaches, our approach not only substantially improves the stem detection accuracy, i.e., distinguishing crop and weed stems, but also provides an improvement in the semantic segmentation performance.