Abstract:Attribution maps are one of the most established tools to explain the functioning of computer vision models. They assign importance scores to input features, indicating how relevant each feature is for the prediction of a deep neural network. While much research has gone into proposing new attribution methods, their proper evaluation remains a difficult challenge. In this work, we propose a novel evaluation protocol that overcomes two fundamental limitations of the widely used incremental-deletion protocol, i.e., the out-of-domain issue and lacking inter-model comparisons. This allows us to evaluate 23 attribution methods and how eight different design choices of popular vision models affect their attribution quality. We find that intrinsically explainable models outperform standard models and that raw attribution values exhibit a higher attribution quality than what is known from previous work. Further, we show consistent changes in the attribution quality when varying the network design, indicating that some standard design choices promote attribution quality.
Abstract:Adapters provide an efficient and lightweight mechanism for adapting trained transformer models to a variety of different tasks. However, they have often been found to be outperformed by other adaptation mechanisms, including low-rank adaptation. In this paper, we provide an in-depth study of adapters, their internal structure, as well as various implementation choices. We uncover pitfalls for using adapters and suggest a concrete, improved adapter architecture, called Adapter+, that not only outperforms previous adapter implementations but surpasses a number of other, more complex adaptation mechanisms in several challenging settings. Despite this, our suggested adapter is highly robust and, unlike previous work, requires little to no manual intervention when addressing a novel scenario. Adapter+ reaches state-of-the-art average accuracy on the VTAB benchmark, even without a per-task hyperparameter optimization.
Abstract:A long-standing challenge in developing machine learning approaches has been the lack of high-quality labeled data. Recently, models trained with purely synthetic data, here termed synthetic clones, generated using large-scale pre-trained diffusion models have shown promising results in overcoming this annotation bottleneck. As these synthetic clone models progress, they are likely to be deployed in challenging real-world settings, yet their suitability remains understudied. Our work addresses this gap by providing the first benchmark for three classes of synthetic clone models, namely supervised, self-supervised, and multi-modal ones, across a range of robustness measures. We show that existing synthetic self-supervised and multi-modal clones are comparable to or outperform state-of-the-art real-image baselines for a range of robustness metrics - shape bias, background bias, calibration, etc. However, we also find that synthetic clones are much more susceptible to adversarial and real-world noise than models trained with real data. To address this, we find that combining both real and synthetic data further increases the robustness, and that the choice of prompt used for generating synthetic images plays an important part in the robustness of synthetic clones.
Abstract:Unsupervised semantic segmentation aims to automatically partition images into semantically meaningful regions by identifying global categories within an image corpus without any form of annotation. Building upon recent advances in self-supervised representation learning, we focus on how to leverage these large pre-trained models for the downstream task of unsupervised segmentation. We present PriMaPs - Principal Mask Proposals - decomposing images into semantically meaningful masks based on their feature representation. This allows us to realize unsupervised semantic segmentation by fitting class prototypes to PriMaPs with a stochastic expectation-maximization algorithm, PriMaPs-EM. Despite its conceptual simplicity, PriMaPs-EM leads to competitive results across various pre-trained backbone models, including DINO and DINOv2, and across datasets, such as Cityscapes, COCO-Stuff, and Potsdam-3. Importantly, PriMaPs-EM is able to boost results when applied orthogonally to current state-of-the-art unsupervised semantic segmentation pipelines.
Abstract:Resource-constrained hardware, such as edge devices or cell phones, often rely on cloud servers to provide the required computational resources for inference in deep vision models. However, transferring image and video data from an edge or mobile device to a cloud server requires coding to deal with network constraints. The use of standardized codecs, such as JPEG or H.264, is prevalent and required to ensure interoperability. This paper aims to examine the implications of employing standardized codecs within deep vision pipelines. We find that using JPEG and H.264 coding significantly deteriorates the accuracy across a broad range of vision tasks and models. For instance, strong compression rates reduce semantic segmentation accuracy by more than 80% in mIoU. In contrast to previous findings, our analysis extends beyond image and action classification to localization and dense prediction tasks, thus providing a more comprehensive perspective.
Abstract:One of the great potentials to improve the confidentiality in mmWave/THz at the physical layer of technical communication, measured by the secrecy rate, lies in the use of reconfigurable intelligent surfaces (RISs). However, an important open problem arises when the eavesdropper is aligned with the legitimate user or in proximity to the RIS or legitimate user. The limitation comes, on one hand, from the high directional gain caused by the dominant line-of-sight (LOS) path in high-frequency transmission, and, on the other hand, from the high energy leakage in the proximity of the RIS and the legitimate user. To address these issues, we employ the concept of frequency diverse arrays (FDA) at the base station (BS) associated with random inverted transmit beamforming and reflective element subset selection (RIBES). More specifically, we consider a passive eavesdropper with unknown location, and design the transmit beamforming and RIS configuration based on the channel information of the legitimate user only. In this context, the secrecy rate with the proposed transmission technique is evaluated in the case of deterministic eavesdropper channel, demonstrating that we can ensure a secure transmission regarding both direction and range. Furthermore, assuming no prior information about the eavesdropper, we describe the wiretap region and derive the worst-case secrecy rate in closed form. The latter is further optimized by determining the optimal subset sizes of the transmit antennas and reflective elements. Simulations verify the correctness of the closed-form expressions and demonstrate that we can effectively improve the secrecy rate, especially when the eavesdropper is close to the RIS or the legitimate user.
Abstract:Automated vehicles operating in urban environments have to reliably interact with other traffic participants. Planning algorithms often utilize separate prediction modules forecasting probabilistic, multi-modal, and interactive behaviors of objects. Designing prediction and planning as two separate modules introduces significant challenges, particularly due to the interdependence of these modules. This work proposes a deep learning methodology to combine prediction and planning. A conditional GAN with the U-Net architecture is trained to predict two high-resolution image sequences. The sequences represent explicit motion predictions, mainly used to train context understanding, and pixel state values suitable for planning encoding kinematic reachability, object dynamics, safety, and driving comfort. The model can be trained offline on target images rendered by a sampling-based model-predictive planner, leveraging real-world driving data. Our results demonstrate intuitive behavior in complex situations, such as lane changes amidst conflicting objectives.
Abstract:Recent years have seen a growing interest in Scene Graph Generation (SGG), a comprehensive visual scene understanding task that aims to predict entity relationships using a relation encoder-decoder pipeline stacked on top of an object encoder-decoder backbone. Unfortunately, current SGG methods suffer from an information loss regarding the entities local-level cues during the relation encoding process. To mitigate this, we introduce the Vision rElation TransfOrmer (VETO), consisting of a novel local-level entity relation encoder. We further observe that many existing SGG methods claim to be unbiased, but are still biased towards either head or tail classes. To overcome this bias, we introduce a Mutually Exclusive ExperT (MEET) learning strategy that captures important relation features without bias towards head or tail classes. Experimental results on the VG and GQA datasets demonstrate that VETO + MEET boosts the predictive performance by up to 47 percentage over the state of the art while being 10 times smaller.
Abstract:The field of explainable artificial intelligence (XAI) aims to uncover the inner workings of complex deep neural models. While being crucial for safety-critical domains, XAI inherently lacks ground-truth explanations, making its automatic evaluation an unsolved problem. We address this challenge by proposing a novel synthetic vision dataset, named FunnyBirds, and accompanying automatic evaluation protocols. Our dataset allows performing semantically meaningful image interventions, e.g., removing individual object parts, which has three important implications. First, it enables analyzing explanations on a part level, which is closer to human comprehension than existing methods that evaluate on a pixel level. Second, by comparing the model output for inputs with removed parts, we can estimate ground-truth part importances that should be reflected in the explanations. Third, by mapping individual explanations into a common space of part importances, we can analyze a variety of different explanation types in a single common framework. Using our tools, we report results for 24 different combinations of neural models and XAI methods, demonstrating the strengths and weaknesses of the assessed methods in a fully automatic and systematic manner.
Abstract:Many convolutional neural networks (CNNs) rely on progressive downsampling of their feature maps to increase the network's receptive field and decrease computational cost. However, this comes at the price of losing granularity in the feature maps, limiting the ability to correctly understand images or recover fine detail in dense prediction tasks. To address this, common practice is to replace the last few downsampling operations in a CNN with dilated convolutions, allowing to retain the feature map resolution without reducing the receptive field, albeit increasing the computational cost. This allows to trade off predictive performance against cost, depending on the output feature resolution. By either regularly downsampling or not downsampling the entire feature map, existing work implicitly treats all regions of the input image and subsequent feature maps as equally important, which generally does not hold. We propose an adaptive downsampling scheme that generalizes the above idea by allowing to process informative regions at a higher resolution than less informative ones. In a variety of experiments, we demonstrate the versatility of our adaptive downsampling strategy and empirically show that it improves the cost-accuracy trade-off of various established CNNs.