We propose a novel semi-supervised active learning framework for monocular 3D object detection with LiDAR guidance (MonoLiG), which leverages all modalities of collected data during model development. We utilize LiDAR to guide the data selection and training of monocular 3D detectors without introducing any overhead in the inference phase. During training, we leverage the LiDAR teacher, monocular student cross-modal framework from semi-supervised learning to distill information from unlabeled data as pseudo-labels. To handle the differences in sensor characteristics, we propose a data noise-based weighting mechanism to reduce the effect of propagating noise from LiDAR modality to monocular. For selecting which samples to label to improve the model performance, we propose a sensor consistency-based selection score that is also coherent with the training objective. Extensive experimental results on KITTI and Waymo datasets verify the effectiveness of our proposed framework. In particular, our selection strategy consistently outperforms state-of-the-art active learning baselines, yielding up to 17% better saving rate in labeling costs. Our training strategy attains the top place in KITTI 3D and bird’s-eye-view (BEV) monocular object detection official benchmarks by improving the BEV Average Precision (AP) by 2.02. Code is shared at https://github.com/aralhekimoglu/monolig.

Multi-task networks demonstrate state-of-the-art performance across various vision tasks. However, their performance relies on large-scale annotated datasets, demanding extensive labeling efforts, especially as the number of tasks to label increases. In this paper, we introduce an active learning framework consisting of a data selection strategy that identifies the most informative unlabeled samples and a training strategy that ensures balanced training across multiple tasks. Our selection strategy leverages the inconsistency between initial and refined task predictions generated by recent two-stage multi-task networks. We further enhance our selection by incorporating task-specific sample diversity through a novel feature extraction mechanism. Our method captures task features for all tasks and distills them into a unified representation, which is used to curate a training set encapsulating diverse task-specific scenarios. In our training strategy, we introduce a sample-specific loss weighting mechanism based on the individual task selection scores. This facilitates the individual prioritization of samples for each task, effectively simulating the sample ordering process inherent in single-task active learning. Extensive experimentation on the PASCAL and NYUD-v2 datasets demonstrates that our approach outperforms existing state-of-the-art methods. Our approach reaches the loss of the network trained with all the available data using only 50% of the data, corresponding to 10% fewer labels compared to the state-of-the-art selection strategy. Our code is available at https://github.com/aralhekimoglu/mtal.

Abstract Since real-world training datasets cannot properly sample the long tail of the underlying data distribution, corner cases and rare out-of-domain samples can severely hinder the performance of state-of-the-art models. This problem becomes even more severe for dense tasks, such as 3D semantic segmentation, where points of non-standard objects can be confidently associated to the wrong class. In this work, we focus on improving the generalization to out-of-domain data. We achieve this by augmenting the training set with adversarial examples. First, we learn a set of vectors that deform the objects in an adversarial fashion. To prevent the adversarial examples from being too far from the existing data distribution, we preserve their plausibility through a series of constraints, ensuring sensor-awareness and shapes smoothness. Then, we perform adversarial augmentation by applying the learned sample-independent vectors to the available objects when training a model. We conduct extensive experiments across a variety of scenarios on data from KITTI, Waymo, and CrashD for 3D object detection, and on data from SemanticKITTI, Waymo, and nuScenes for 3D semantic segmentation. Despite training on a standard single dataset, our approach substantially improves the robustness and generalization of both 3D object detection and 3D semantic segmentation methods to out-of-domain data.

Panoptic segmentation methods assign a known class to each pixel given in input. Even for state-of-the-art approaches, this inevitably enforces decisions that systematically lead to wrong predictions for objects outside the training categories. However, robustness against out-of-distribution samples and corner cases is crucial in safety-critical settings to avoid dangerous consequences. Since real-world datasets cannot contain enough data points to adequately sample the long tail of the underlying distribution, models must be able to deal with unseen and unknown scenarios as well. Previous methods targeted this by re-identifying already-seen unlabeled objects. In this work, we propose the necessary step to extend segmentation with a new setting which we term holistic segmentation. Holistic segmentation aims to identify and separate objects of unseen, unknown categories into instances without any prior knowledge about them while performing panoptic segmentation of known classes. We tackle this new problem with U3HS, which finds unknowns as highly uncertain regions and clusters their corresponding instance-aware embeddings into individual objects. By doing so, for the first time in panoptic segmentation with unknown objects, our U3HS is trained without unknown categories, reducing assumptions and leaving the settings as unconstrained as in real-life scenarios. Extensive experiments on public data from MS COCO, Cityscapes, and Lost&Found demonstrate the effectiveness of U3HS for this new, challenging, and assumptions-free setting called holistic segmentation. Project page: https://holisticseg.github.io.

As 3D object detection on point clouds relies on the geometrical relationships between the points, non-standard object shapes can hinder a method's detection capability. However, in safety-critical settings, robustness to out-of-domain and long-tail samples is fundamental to circumvent dangerous issues, such as the misdetection of damaged or rare cars. In this work, we substantially improve the generalization of 3D object detectors to out-of-domain data by deforming point clouds during training. We achieve this with 3D-VField: a novel data augmentation method that plausibly deforms objects via vector fields learned in an adversarial fashion. Our approach constrains 3D points to slide along their sensor view rays while neither adding nor removing any of them. The obtained vectors are transferable, sample-independent and preserve shape and occlusions. Despite training only on a standard dataset, such as KITTI, augmenting with our vector fields significantly improves the generalization to differently shaped objects and scenes. Towards this end, we propose and share CrashD: a synthetic dataset of realistic damaged and rare cars, with a variety of crash scenarios. Extensive experiments on KITTI, Waymo, our CrashD and SUN RGB-D show the generalizability of our techniques to out-of-domain data, different models and sensors, namely LiDAR and ToF cameras, for both indoor and outdoor scenes. Our CrashD dataset is available at https://crashd-cars.github.io.

Processing camera information to perceive their 3D surrounding is essential for building scalable autonomous driving vehicles. For this task, deep learning networks provide effective real-time solutions. However, to compensate for missing depth information in cameras compared to LiDARs, a large amount of labeled data is required for training. Active learning is a training framework where the network actively participates in the data selection process to improve data efficiency and performance. In this work, we propose an active learning pipeline for 3D object detection from monocular images. The main components of our approach are (1) two training-efficient uncertainty estimation strategies, (2) a diversity-based selection strategy to select images that contain the most diverse set of objects, (3) a novel active learning strategy more suitable for training autonomous driving perception networks. Experiments show that combining our proposed uncertainty estimation methods provides a better data saving rate and reaches a higher final performance than baselines. Furthermore, we empirically show performance gains of the presented diversity-based selection strategy and the efficiency of the proposed active learning strategy.

Employment interviews are relevant scenarios for the study of social interaction. In this setting, social skills play an important role, even though the interactions between potential employers and candidates are often limited. One fundamental aspect of social interaction is the use of nonverbal communication , which affects how we are socially perceived. We present a method to automatically extract body communicative cues from one-on-one conversations recorded with Kinect devices. First, we find the three-dimensional position of hands and head of the subject, and, aided by training data, we infer the upper body pose. Then, we use the inferred poses to perform action recognition and build person-specific activity descriptors. We evaluate our system with both domain-specific and public, generic datasets, and show competitive performance.

We address the problem of body communication retrieval and measuring in seated conversations by means of markerless motion capture. In psychological studies, the use of automatic methods is key to reduce the subjectivity present in manual behavioral coding used to extract these cues. These studies usually involve hundreds of subjects with different clothing, non-acted poses, or different distances to the camera in uncalibrated, RGB-only video. However, range cameras are not yet common in psychology research, especially in existing recordings. Therefore, it becomes highly relevant to develop a fast method that is able to work in these conditions. Given the known relationship between depth and motion estimates, we propose to robustly integrate highly appearance-invariant image motion features in a machine learning approach, complemented with an effective tracking scheme. We evaluate the method's performance with existing databases and a database of upper body poses displayed in job interviews that we make public, showing that in our scenario it is comparable to that of Kinect without using a range camera, and state-of-the-art w.r.t. the HumanEva and ChaLearn 2011 evaluation datasets.

We present a robust method to automatically detect blinks in video sequences of conversations, aimed to discovering stress. Psychological studies have shown a relationship between blink frequency and dopamine levels, which in turn are affected by stress. Task performance correlates through an inverted U shape to both dopamine and stress levels. This shows the importance of automatic blink detection as a way of reducing human coding burden. We use an off-the-shelf face tracker in order to extract the eye region. Then, we perform per-pixel classification of the extracted eye images to later identify blinks through their dynamics. We evaluate the performance of our system with a job interview database with annotations of psychological variables, and show statistically significant correlation between perceived stress resistance and the automatically detected blink patterns.

Hand gestures and body posture are intimately linked to speech as they are used to enrich the vocal content, and are therefore inherently multimodal. As an important part of nonverbal behavior, body communication carries relevant information that can reveal social constructs as diverse as personality, internal states, or job interview outcomes. In this work, we analyze body communication cues in real dyadic employment interviews, where the protagonists of the interaction are seated. We use a mixture of body communicative features based on manual annotations and automated extraction methods to successfully predict two key organizational constructs, namely personality and job interview ratings. Our work also confirms the multimodal nature of body communication and shows that the speaking status can be used to improve the prediction performance of personality and hirability.