Most active learning research has focused on methods which perform well when many labels are available, but can be dramatically worse than random selection when label budgets are small. Other methods have focused on the low-budget regime, but do poorly as label budgets increase. As the line between "low" and "high" budgets varies by problem, this is a serious issue in practice. We propose *uncertainty coverage*, an objective which generalizes a variety of low- and high-budget objectives, as well as natural, hyperparameter-light methods to smoothly interpolate between low- and high-budget regimes. We call greedy optimization of the estimate Uncertainty Herding; this simple method is computationally fast, and we prove that it nearly optimizes the distribution-level coverage. In experimental validation across a variety of active learning tasks, our proposal matches or beats state-of-the-art performance in essentially all cases; it is the only method of which we are aware that reliably works well in both low- and high-budget settings.

Despite attaining high empirical generalization, the sharpness of models trained with sharpness-aware minimization (SAM) do not always correlate with generalization error. Instead of viewing SAM as minimizing sharpness to improve generalization, our paper considers a new perspective based on SAM’s training dynamics. We propose that perturbations in SAM perform perturbed forgetting, where they discard undesirable model biases to exhibit learning signals that generalize better. We relate our notion of forgetting to the information bottleneck principle, use it to explain observations like the better generalization of smaller perturbation batches, and show that perturbed forgetting can exhibit a stronger correlation with generalization than flatness. While standard SAM targets model biases exposed by the steepest ascent directions, we propose a new perturbation that targets biases exposed through the model’s outputs. Our output bias forgetting perturbations outperform standard SAM, GSAM, and ASAM on ImageNet, robustness benchmarks, and transfer to CIFAR-10,100, while sometimes converging to sharper regions. Our results suggest that the benefits of SAM can be explained by alternative mechanistic principles that do not require flatness of the loss surface.

Predicting multiple heterogeneous biological and medical targets is a challenge for traditional deep learning models. In contrast to single-task learning, in which a separate model is trained for each target, multi-task learning (MTL) optimizes a single model to predict multiple related targets simultaneously. To address this challenge, we propose the Multi-gate Mixture-of-Experts with Exclusivity (MMoEEx). Our work aims to tackle the heterogeneous MTL setting, in which the same model optimizes multiple tasks with different characteristics. Such a scenario can overwhelm current MTL approaches due to the challenges in balancing shared and task-specific representations and the need to optimize tasks with competing optimization paths. Our method makes two key contributions: first, we introduce an approach to induce more diversity among experts, thus creating representations more suitable for highly imbalanced and heterogenous MTL learning; second, we adopt a two-step optimization (Finn et al., 2017 and Lee et al., 2020) approach to balancing the tasks at the gradient level. We validate our method on three MTL benchmark datasets, including UCI-Census-income dataset, Medical Information Mart for Intensive Care (MIMIC-III) and PubChem BioAssay (PCBA).

Densely connected networks for classification enable feature exploration and result in state-of-the-art performance on multiple classification tasks. The alternative to dense networks is the residual network which enables feature re-usage. In this work, we combine these orthogonal concepts for encoder-decoder architectures, which we call Dual-Path Dense-Block Network (DPDB-Net). We introduce a dense block which incorporates feature re-usage and new feature exploration in the encoder. Moreover, we discuss that feature re-usage by the residual network architecture leads to a feature map explosion in the decoder and, thus, is not advantageous in this part of the network. We evaluated our proposed architecture in multiple segmentation tasks and report state-of-the-art performance on the Freiburg Forest dataset and competitive results on the Cam Vid dataset.

General human action recognition requires understanding of various visual cues. In this paper, we propose a network architecture that computes and integrates the most important visual cues for action recognition: pose, motion, and the raw images. For the integration, we introduce a Markov chain model which adds cues successively. The resulting approach is efficient and applicable to action classification as well as to spatial and temporal action localization. The two contributions clearly improve the performance over respective baselines. The overall approach achieves state-of-the-art action classification performance on HMDB51, J-HMDB and NTU RGB+D datasets. Moreover, it yields state-of-the-art spatio-temporal action localization results on UCF101 and J-HMDB.

Visual place recognition under difficult perceptual conditions remains a challenging problem due to changing weather conditions, illumination and seasons. Long-term visual navigation approaches for robot localization should be robust to these dynamics of the environment. Existing methods typically leverage feature descriptions of whole images or image regions from Deep Convolutional Neural Networks. Some approaches also exploit sequential information to alleviate the problem of spatially inconsistent and non-perfect image matches. In this paper, we propose a novel approach for learning a discriminative holistic image representation which exploits the image content to create a dense and salient scene description. These salient descriptions are learnt over a variety of datasets under large perceptual changes. Such an approach enables us to precisely segment the regions of an image which are geometrically stable over large time lags. We combine features from these salient regions and an off-the-shelf holistic representation to form a more robust scene descriptor. We also introduce a semantically labeled dataset which captures extreme perceptual and structural scene dynamics over the course of 3 years. We evaluated our approach with extensive experiments on data collected over several kilometers in Freiburg and show that our learnt image representation outperforms off-the-shelf features from the deep networks and hand-crafted features.

Robots are expected to operate autonomously in dynamic environments. Understanding the underlying dynamic characteristics of objects is a key enabler for achieving this goal. In this paper, we propose a method for pointwise semantic classification of 3D LiDAR data into three classes: non-movable, movable and dynamic. We concentrate on understanding these specific semantics because they characterize important information required for an autonomous system. To learn the distinction between movable and non-movable points in the environment, we introduce an approach based on deep neural network and for detecting the dynamic points, we estimate pointwise motion. We propose a Bayes filter framework for combining the learned semantic cues with the motion cues to infer the required semantic classification. In extensive experiments, we compare our approach with other methods on a standard benchmark dataset and report competitive results in comparison to the existing state-of-the-art. Furthermore, we show an improvement in the classification of points by combining the semantic cues retrieved from the neural network with the motion cues.

This paper addresses the problem of human body part segmentation in conventional RGB images, which has several applications in robotics, such as learning from demonstration and human-robot handovers. The proposed solution is based on Convolutional Neural Networks (CNNs). We present a network architecture that assigns each pixel to one of a predefined set of human body part classes, such as head, torso, arms, legs. After initializing weights with a very deep convolutional network for image classification, the network can be trained end-to-end and yields precise class predictions at the original input resolution. Our architecture particularly improves on over-fitting issues in the up-convolutional part of the network. Relying only on RGB rather than RGB-D images also allows us to apply the approach outdoors. The network achieves state-of-the-art performance on the PASCAL Parts dataset. Moreover, we introduce two new part segmentation datasets, the Freiburg sitting people dataset and the Freiburg people in disaster dataset. We also present results obtained with a ground robot and an unmanned aerial vehicle.

This paper addresses the problem of road scene segmentation in conventional RGB images by exploiting recent advances in semantic segmentation via convolutional neural networks (CNNs). Segmentation networks are very large and do not currently run at interactive frame rates. To make this technique applicable to robotics we propose several architecture refinements that provide the best trade-off between segmentation quality and runtime. This is achieved by a new mapping between classes and filters at the expansion side of the network. The network is trained end-to-end and yields precise road/lane predictions at the original input resolution in roughly 50ms. Compared to the state of the art, the network achieves top accuracies on the KITTI dataset for road and lane segmentation while providing a 20× speed-up. We demonstrate that the improved efficiency is not due to the road segmentation task. Also on segmentation datasets with larger scene complexity, the accuracy does not suffer from the large speed-up.

Successful image-based object recognition techniques have been constructed founded on powerful techniques such as sparse representation, in lieu of the popular vector quantization approach. However, one serious drawback of sparse space-based methods is that local features that are quite similar can be quantized into quite distinct visual words. We address this problem with a novel approach for object recognition, called sparse spatial coding, which efficiently combines a sparse coding dictionary learning and spatial constraint coding stage. We performed experimental evaluation using the Caltech 101, Caltech 256, Corel 5000, and Corel 10000 data sets, which were specifically designed for object recognition evaluation. Our results show that our approach achieves high accuracy comparable with the best single feature method previously published on those databases. Our method outperformed, for the same bases, several multiple feature methods, and provided equivalent, and in few cases, slightly less accurate results than other techniques specifically designed to that end. Finally, we report state-of-the-art results for scene recognition on COsy Localization Dataset (COLD) and high performance results on the MIT-67 indoor scene recognition, thus demonstrating the generalization of our approach for such tasks.

Successful state-of-the-art object recognition techniques from images have been based on powerful methods, such as sparse representation, in order to replace the also popular vector quantization (VQ) approach. Recently, sparse coding, which is characterized by representing a signal in a sparse space, has raised the bar on several object recognition benchmarks. However, one serious drawback of sparse space based methods is that similar local features can be quantized into different visual words. We present in this paper a new method, called Sparse Spatial Coding (SSC), which combines a sparse coding dictionary learning, a spatial constraint coding stage and an online classification method to improve object recognition. An efficient new off-line classification algorithm is also presented. We overcome the problem of techniques which make use of sparse representation alone by generating the final representation with SSC and max pooling, presented for an online learning classifier. Experimental results obtained on the Caltech 101, Caltech 256, Corel 5000 and Corel 10000 databases, show that, to the best of our knowledge, our approach supersedes in accuracy the best published results to date on the same databases. As an extension, we also show high performance results on the MIT-67 indoor scene recognition dataset.

This work introduces a novel descriptor called Binary Robust Appearance and Normals Descriptor (BRAND), that efficiently combines appearance and geometric shape information from RGB-D images, and is largely invariant to rotation and scale transform. The proposed approach encodes point information as a binary string providing a descriptor that is suitable for applications that demand speed performance and low memory consumption. Results of several experiments demonstrate that as far as precision and robustness are concerned, BRAND achieves improved results when compared to state of the art descriptors based on texture, geometry and combination of both information. We also demonstrate that our descriptor is robust and provides reliable results in a registration task even when a sparsely textured and poorly illuminated scene is used.