New advances in universal approximation with neural networks of minimal width

Deep neural networks have achieved remarkable success in diverse applications, prompting the need for a solid theoretical foundation. Recent research has identified the minimal width $\max\{2,d_x,d_y\}$ required for neural networks with input dimensions $d_x$ and output dimension $d_y$ that use leaky ReLU activations to universally approximate $L^p(\mathbb{R}^{d_x},\mathbb{R}^{d_y})$ on compacta. Here, we present an alternative proof for the minimal width of such neural networks, by directly constructing approximating networks using a coding scheme that leverages the properties of leaky ReLUs and standard $L^p$ results. The obtained construction has a minimal interior dimension of $1$, independent of input and output dimensions, which allows us to show that autoencoders with leaky ReLU activations are universal approximators of $L^p$ functions. Furthermore, we demonstrate that the normalizing flow LU-Net serves as a distributional universal approximator. We broaden our results to show that smooth invertible neural networks can approximate $L^p(\mathbb{R}^{d},\mathbb{R}^{d})$ on compacta when the dimension $d\geq 2$, which provides a constructive proof of a classical theorem of Brenier and Gangbo. In addition, we use a topological argument to establish that for FNNs with monotone Lipschitz continuous activations, $d_x+1$ is a lower bound on the minimal width required for the uniform universal approximation of continuous functions $C^0(\mathbb{R}^{d_x},\mathbb{R}^{d_y})$ on compacta when $d_x\geq d_y$.

On the Influence of Shape, Texture and Color for Learning Semantic Segmentation

In recent years, a body of works has emerged, studying shape and texture biases of off-the-shelf pre-trained deep neural networks (DNN) for image classification. These works study how much a trained DNN relies on image cues, predominantly shape and texture. In this work, we switch the perspective, posing the following questions: What can a DNN learn from each of the image cues, i.e., shape, texture and color, respectively? How much does each cue influence the learning success? And what are the synergy effects between different cues? Studying these questions sheds light upon cue influences on learning and thus the learning capabilities of DNNs. We study these questions on semantic segmentation which allows us to address our questions on pixel level. To conduct this study, we develop a generic procedure to decompose a given dataset into multiple ones, each of them only containing either a single cue or a chosen mixture. This framework is then applied to two real-world datasets, Cityscapes and PASCAL Context, and a synthetic data set based on the CARLA simulator. We learn the given semantic segmentation task from these cue datasets, creating cue experts. Early fusion of cues is performed by constructing appropriate datasets. This is complemented by a late fusion of experts which allows us to study cue influence location-dependent on pixel level. Our study on three datasets reveals that neither texture nor shape clearly dominate the learning success, however a combination of shape and color but without texture achieves surprisingly strong results. Our findings hold for convolutional and transformer backbones. In particular, qualitatively there is almost no difference in how both of the architecture types extract information from the different cues.

VL4AD: Vision-Language Models Improve Pixel-wise Anomaly Detection

Semantic segmentation networks have achieved significant success under the assumption of independent and identically distributed data. However, these networks often struggle to detect anomalies from unknown semantic classes due to the limited set of visual concepts they are typically trained on. To address this issue, anomaly segmentation often involves fine-tuning on outlier samples, necessitating additional efforts for data collection, labeling, and model retraining. Seeking to avoid this cumbersome work, we take a different approach and propose to incorporate Vision-Language (VL) encoders into existing anomaly detectors to leverage the semantically broad VL pre-training for improved outlier awareness. Additionally, we propose a new scoring function that enables data- and training-free outlier supervision via textual prompts. The resulting VL4AD model, which includes max-logit prompt ensembling and a class-merging strategy, achieves competitive performance on widely used benchmark datasets, thereby demonstrating the potential of vision-language models for pixel-wise anomaly detection.

Hyperparameter Optimization for Driving Strategies Based on Reinforcement Learning

This paper focuses on hyperparameter optimization for autonomous driving strategies based on Reinforcement Learning. We provide a detailed description of training the RL agent in a simulation environment. Subsequently, we employ Efficient Global Optimization algorithm that uses Gaussian Process fitting for hyperparameter optimization in RL. Before this optimization phase, Gaussian process interpolation is applied to fit the surrogate model, for which the hyperparameter set is generated using Latin hypercube sampling. To accelerate the evaluation, parallelization techniques are employed. Following the hyperparameter optimization procedure, a set of hyperparameters is identified, resulting in a noteworthy enhancement in overall driving performance. There is a substantial increase of 4\% when compared to existing manually tuned parameters and the hyperparameters discovered during the initialization process using Latin hypercube sampling. After the optimization, we analyze the obtained results thoroughly and conduct a sensitivity analysis to assess the robustness and generalization capabilities of the learned autonomous driving strategies. The findings from this study contribute to the advancement of Gaussian process based Bayesian optimization to optimize the hyperparameters for autonomous driving in RL, providing valuable insights for the development of efficient and reliable autonomous driving systems.

MetaToken: Detecting Hallucination in Image Descriptions by Meta Classification

Large Vision Language Models (LVLMs) have shown remarkable capabilities in multimodal tasks like visual question answering or image captioning. However, inconsistencies between the visual information and the generated text, a phenomenon referred to as hallucinations, remain an unsolved problem with regard to the trustworthiness of LVLMs. To address this problem, recent works proposed to incorporate computationally costly Large (Vision) Language Models in order to detect hallucinations on a sentence- or subsentence-level. In this work, we introduce MetaToken, a lightweight binary classifier to detect hallucinations on the token-level at negligible cost. Based on a statistical analysis, we reveal key factors of hallucinations in LVLMs which have been overseen in previous works. MetaToken can be applied to any open-source LVLM without any knowledge about ground truth data providing a reliable detection of hallucinations. We evaluate our method on four state-of-the-art LVLMs demonstrating the effectiveness of our approach.

AnoVox: A Benchmark for Multimodal Anomaly Detection in Autonomous Driving

The scale-up of autonomous vehicles depends heavily on their ability to deal with anomalies, such as rare objects on the road. In order to handle such situations, it is necessary to detect anomalies in the first place. Anomaly detection for autonomous driving has made great progress in the past years but suffers from poorly designed benchmarks with a strong focus on camera data. In this work, we propose AnoVox, the largest benchmark for ANOmaly detection in autonomous driving to date. AnoVox incorporates large-scale multimodal sensor data and spatial VOXel ground truth, allowing for the comparison of methods independent of their used sensor. We propose a formal definition of normality and provide a compliant training dataset. AnoVox is the first benchmark to contain both content and temporal anomalies.

VLTSeg: Simple Transfer of CLIP-Based Vision-Language Representations for Domain Generalized Semantic Segmentation

Domain generalization (DG) remains a significant challenge for perception based on deep neural networks (DNN), where domain shifts occur due to lighting, weather, or geolocation changes. In this work, we propose VLTSeg to enhance domain generalization in semantic segmentation, where the network is solely trained on the source domain and evaluated on unseen target domains. Our method leverages the inherent semantic robustness of vision-language models. First, by substituting traditional vision-only backbones with pre-trained encoders from CLIP and EVA-CLIP as transfer learning setting we find that in the field of DG, vision-language pre-training significantly outperforms supervised and self-supervised vision pre-training. We thus propose a new vision-language approach for domain generalized segmentation, which improves the domain generalization SOTA by 7.6% mIoU when training on the synthetic GTA5 dataset. We further show the superior generalization capabilities of vision-language segmentation models by reaching 76.48% mIoU on the popular Cityscapes-to-ACDC benchmark, outperforming the previous SOTA approach by 6.9% mIoU on the test set at the time of writing. Additionally, our approach shows strong in-domain generalization capabilities indicated by 86.1% mIoU on the Cityscapes test set, resulting in a shared first place with the previous SOTA on the current leaderboard at the time of submission.

Temporal Performance Prediction for Deep Convolutional Long Short-Term Memory Networks

Quantifying predictive uncertainty of deep semantic segmentation networks is essential in safety-critical tasks. In applications like autonomous driving, where video data is available, convolutional long short-term memory networks are capable of not only providing semantic segmentations but also predicting the segmentations of the next timesteps. These models use cell states to broadcast information from previous data by taking a time series of inputs to predict one or even further steps into the future. We present a temporal postprocessing method which estimates the prediction performance of convolutional long short-term memory networks by either predicting the intersection over union of predicted and ground truth segments or classifying between intersection over union being equal to zero or greater than zero. To this end, we create temporal cell state-based input metrics per segment and investigate different models for the estimation of the predictive quality based on these metrics. We further study the influence of the number of considered cell states for the proposed metrics.

Have We Ever Encountered This Before? Retrieving Out-of-Distribution Road Obstacles from Driving Scenes

In the life cycle of highly automated systems operating in an open and dynamic environment, the ability to adjust to emerging challenges is crucial. For systems integrating data-driven AI-based components, rapid responses to deployment issues require fast access to related data for testing and reconfiguration. In the context of automated driving, this especially applies to road obstacles that were not included in the training data, commonly referred to as out-of-distribution (OoD) road obstacles. Given the availability of large uncurated recordings of driving scenes, a pragmatic approach is to query a database to retrieve similar scenarios featuring the same safety concerns due to OoD road obstacles. In this work, we extend beyond identifying OoD road obstacles in video streams and offer a comprehensive approach to extract sequences of OoD road obstacles using text queries, thereby proposing a way of curating a collection of OoD data for subsequent analysis. Our proposed method leverages the recent advances in OoD segmentation and multi-modal foundation models to identify and efficiently extract safety-relevant scenes from unlabeled videos. We present a first approach for the novel task of text-based OoD object retrieval, which addresses the question ''Have we ever encountered this before?''.

ResBuilder: Automated Learning of Depth with Residual Structures

In this work, we develop a neural architecture search algorithm, termed Resbuilder, that develops ResNet architectures from scratch that achieve high accuracy at moderate computational cost. It can also be used to modify existing architectures and has the capability to remove and insert ResNet blocks, in this way searching for suitable architectures in the space of ResNet architectures. In our experiments on different image classification datasets, Resbuilder achieves close to state-of-the-art performance while saving computational cost compared to off-the-shelf ResNets. Noteworthy, we once tune the parameters on CIFAR10 which yields a suitable default choice for all other datasets. We demonstrate that this property generalizes even to industrial applications by applying our method with default parameters on a proprietary fraud detection dataset.