Datasets are essential for training and testing vehicle perception algorithms. However, the collection and annotation of real-world images is time-consuming and expensive. Driving simulators offer a solution by automatically generating various driving scenarios with corresponding annotations, but the simulation-to-reality (Sim2Real) domain gap remains a challenge. While most of the Generative Artificial Intelligence (AI) follows the de facto Generative Adversarial Nets (GANs)-based methods, the recent emerging diffusion probabilistic models have not been fully explored in mitigating Sim2Real challenges for driving data synthesis. To explore the performance, this paper applied three different generative AI methods to leverage semantic label maps from a driving simulator as a bridge for the creation of realistic datasets. A comparative analysis of these methods is presented from the perspective of image quality and perception. New synthetic datasets, which include driving images and auto-generated high-quality annotations, are produced with low costs and high scene variability. The experimental results show that although GAN-based methods are adept at generating high-quality images when provided with manually annotated labels, ControlNet produces synthetic datasets with fewer artefacts and more structural fidelity when using simulator-generated labels. This suggests that the diffusion-based approach may provide improved stability and an alternative method for addressing Sim2Real challenges.
Diffusion Models (DMs) have achieved State-Of-The-Art (SOTA) results in the Lidar point cloud generation task, benefiting from their stable training and iterative refinement during sampling. However, DMs often fail to realistically model Lidar raydrop noise due to their inherent denoising process. To retain the strength of iterative sampling while enhancing the generation of raydrop noise, we introduce LidarGRIT, a generative model that uses auto-regressive transformers to iteratively sample the range images in the latent space rather than image space. Furthermore, LidarGRIT utilises VQ-VAE to separately decode range images and raydrop masks. Our results show that LidarGRIT achieves superior performance compared to SOTA models on KITTI-360 and KITTI odometry datasets. Code available at:https://github.com/hamedhaghighi/LidarGRIT.
Precise situational awareness is required for the safe decision-making of assisted and automated driving (AAD) functions. Panoptic segmentation is a promising perception technique to identify and categorise objects, impending hazards, and driveable space at a pixel level. While segmentation quality is generally associated with the quality of the camera data, a comprehensive understanding and modelling of this relationship are paramount for AAD system designers. Motivated by such a need, this work proposes a unifying pipeline to assess the robustness of panoptic segmentation models for AAD, correlating it with traditional image quality. The first step of the proposed pipeline involves generating degraded camera data that reflects real-world noise factors. To this end, 19 noise factors have been identified and implemented with 3 severity levels. Of these factors, this work proposes novel models for unfavourable light and snow. After applying the degradation models, three state-of-the-art CNN- and vision transformers (ViT)-based panoptic segmentation networks are used to analyse their robustness. The variations of the segmentation performance are then correlated to 8 selected image quality metrics. This research reveals that: 1) certain specific noise factors produce the highest impact on panoptic segmentation, i.e. droplets on lens and Gaussian noise; 2) the ViT-based panoptic segmentation backbones show better robustness to the considered noise factors; 3) some image quality metrics (i.e. LPIPS and CW-SSIM) correlate strongly with panoptic segmentation performance and therefore they can be used as predictive metrics for network performance.
Perception sensor models are essential elements of automotive simulation environments; they also serve as powerful tools for creating synthetic datasets to train deep learning-based perception models. Developing realistic perception sensor models poses a significant challenge due to the large gap between simulated sensor data and real-world sensor outputs, known as the sim-to-real gap. To address this problem, learning-based models have emerged as promising solutions in recent years, with unparalleled potential to map low-fidelity simulated sensor data into highly realistic outputs. Motivated by this potential, this paper focuses on sim-to-real mapping of Lidar point clouds, a widely used perception sensor in automated driving systems. We introduce a novel Contrastive-Learning-based Sim-to-Real mapping framework, namely CLS2R, inspired by the recent advancements in image-to-image translation techniques. The proposed CLS2R framework employs a lossless representation of Lidar point clouds, considering all essential Lidar attributes such as depth, reflectance, and raydrop. We extensively evaluate the proposed framework, comparing it with state-of-the-art image-to-image translation methods using a diverse range of metrics to assess realness, faithfulness, and the impact on the performance of a downstream task. Our results show that CLS2R demonstrates superior performance across nearly all metrics. Source code is available at https://github.com/hamedhaghighi/CLS2R.git.
Due to the costliness of labelled data in real-world applications, semi-supervised learning, underpinned by pseudo labelling, is an appealing solution. However, handling confusing samples is nontrivial: discarding valuable confusing samples would compromise the model generalisation while using them for training would exacerbate the issue of confirmation bias caused by the resulting inevitable mislabelling. To solve this problem, this paper proposes to use confusing samples proactively without label correction. Specifically, a Virtual Category (VC) is assigned to each confusing sample in such a way that it can safely contribute to the model optimisation even without a concrete label. This provides an upper bound for inter-class information sharing capacity, which eventually leads to a better embedding space. Extensive experiments on two mainstream dense prediction tasks -- semantic segmentation and object detection, demonstrate that the proposed VC learning significantly surpasses the state-of-the-art, especially when only very few labels are available. Our intriguing findings highlight the usage of VC learning in dense vision tasks.
A CF explainer identifies the minimum modifications in the input that would alter the model's output to its complement. In other words, a CF explainer computes the minimum modifications required to cross the model's decision boundary. Current deep generative CF models often work with user-selected features rather than focusing on the discriminative features of the black-box model. Consequently, such CF examples may not necessarily lie near the decision boundary, thereby contradicting the definition of CFs. To address this issue, we propose in this paper a novel approach that leverages saliency maps to generate more informative CF explanations. Source codes are available at: https://github.com/Amir-Samadi//Saliency_Aware_CF.
Due to the costliness of labelled data in real-world applications, semi-supervised object detectors, underpinned by pseudo labelling, are appealing. However, handling confusing samples is nontrivial: discarding valuable confusing samples would compromise the model generalisation while using them for training would exacerbate the confirmation bias issue caused by inevitable mislabelling. To solve this problem, this paper proposes to use confusing samples proactively without label correction. Specifically, a virtual category (VC) is assigned to each confusing sample such that they can safely contribute to the model optimisation even without a concrete label. It is attributed to specifying the embedding distance between the training sample and the virtual category as the lower bound of the inter-class distance. Moreover, we also modify the localisation loss to allow high-quality boundaries for location regression. Extensive experiments demonstrate that the proposed VC learning significantly surpasses the state-of-the-art, especially with small amounts of available labels.
Recently, Deep Learning-based methods for inverse tone-mapping standard dynamic range (SDR) images to obtain high dynamic range (HDR) images have become very popular. These methods manage to fill over-exposed areas convincingly both in terms of details and dynamic range. Typically, these methods, to be effective, need to learn from large datasets and to transfer this knowledge to the network weights. In this work, we tackle this problem from a completely different perspective. What can we learn from a single SDR video? With the presented zero-shot approach, we show that, in many cases, a single SDR video is sufficient to be able to generate an HDR video of the same quality or better than other state-of-the-art methods.
Inverse Tone Mapping (ITM) methods attempt to reconstruct High Dynamic Range (HDR) information from Low Dynamic Range (LDR) image content. The dynamic range of well-exposed areas must be expanded and any missing information due to over/under-exposure must be recovered (hallucinated). The majority of methods focus on the former and are relatively successful, while most attempts on the latter are not of sufficient quality, even ones based on Convolutional Neural Networks (CNNs). A major factor for the reduced inpainting quality in some works is the choice of loss function. Work based on Generative Adversarial Networks (GANs) shows promising results for image synthesis and LDR inpainting, suggesting that GAN losses can improve inverse tone mapping results. This work presents a GAN-based method that hallucinates missing information from badly exposed areas in LDR images and compares its efficacy with alternative variations. The proposed method is quantitatively competitive with state-of-the-art inverse tone mapping methods, providing good dynamic range expansion for well-exposed areas and plausible hallucinations for saturated and under-exposed areas. A density-based normalisation method, targeted for HDR content, is also proposed, as well as an HDR data augmentation method targeted for HDR hallucination.
Dual-panel displays require local dimming algorithms in order to reproduce content with high fidelity and high dynamic range. In this work, a novel deep learning based local dimming method is proposed for rendering HDR images on dual-panel HDR displays. The method uses a Convolutional Neural Network to predict backlight values, using as input the HDR image that is to be displayed. The model is designed and trained via a controllable power parameter that allows a user to trade off between power and quality. The proposed method is evaluated against six other methods on a test set of 105 HDR images, using a variety of quantitative quality metrics. Results demonstrate improved display quality and better power consumption when using the proposed method compared to the best alternatives.