Corpus ID: 208637360

Let's Get Dirty: GAN Based Data Augmentation for Soiling and Adverse Weather Classification in Autonomous Driving

  title={Let's Get Dirty: GAN Based Data Augmentation for Soiling and Adverse Weather Classification in Autonomous Driving},
  author={Michal Uři{\vc}{\'a}ř and Ganesh Sistu and Hazem Rashed and Anton{\'i}n Vobeck{\'y} and Pavel Kr{\'i}zek and Fabian Burger and Senthil Kumar Yogamani},
Cameras are getting more and more important in autonomous driving. Wide-angle fisheye cameras are relatively cheap sensors and very suitable for automated parking and low-speed navigation tasks. Four of such cameras form a surround-view system that provides a complete and detailed view around the vehicle. These cameras are usually directly exposed to harsh environmental settings and therefore can get soiled very easily by mud, dust, water, frost, etc. The soiling on the camera lens has a direct… Expand
Model-Based Occlusion Disentanglement for Image-to-Image Translation
The experiments demonstrate the unsupervised model-based learning method is able to handle varying types of occlusions and generate highly realistic translations, qualitatively and quantitatively outperforming the state of the art on multiple datasets. Expand
Guided Disentanglement in Generative Networks
This paper presents a comprehensive method for disentangling physics-based traits in the translation, guiding the learning process with neural or physical models and integrating adversarial estimation and genetic algorithms to correctly achieve disentanglement. Expand
Model-based disentanglement of lens occlusions
An unsupervised model-based disentanglement training, which learns to disentangle scene from lens occlusions and can regress the occlusion model parameters from target database, is proposed. Expand


Desoiling Dataset: Restoring Soiled Areas on Automotive Fisheye Cameras
A dataset for measuring the performance degradation as well as to help constructing classifiers for soiling detection, or for trying to restore the soiled images, so to increase the performance of the off-the-shelf classifiers. Expand
SoilingNet: Soiling Detection on Automotive Surround-View Cameras
This work created a new dataset having multiple types of soiling namely opaque and transparent and demonstrates high accuracy using a Convolutional Neural Network (CNN) based architecture and makes use of Generative Adversarial Networks to generate more images for data augmentation. Expand
FisheyeMultiNet: Real-time Multi-task Learning Architecture for Surround-view Automated Parking System
A holistic overview of an industrial system covering the embedded system, use cases and the deep learning architecture is provided, including a real-time multi-task deep learning network called FisheyeMultiNet, which detects all the necessary objects for parking on a low-power embedded system. Expand
WoodScape: A Multi-Task, Multi-Camera Fisheye Dataset for Autonomous Driving
The first extensive fisheye automotive dataset, WoodScape, named after Robert Wood, which comprises of four surround view cameras and nine tasks including segmentation, depth estimation, 3D bounding box detection and soiling detection is released. Expand
Spatial Attentive Single-Image Deraining With a High Quality Real Rain Dataset
A semi-automatic method that incorporates temporal priors and human supervision to generate a high-quality clean image from each input sequence of real rain images is proposed, and a novel SPatial Attentive Network (SPANet) is proposed to remove rain streaks in a local-to-global manner. Expand
The Cityscapes Dataset for Semantic Urban Scene Understanding
This work introduces Cityscapes, a benchmark suite and large-scale dataset to train and test approaches for pixel-level and instance-level semantic labeling, and exceeds previous attempts in terms of dataset size, annotation richness, scene variability, and complexity. Expand
MODNet: Motion and Appearance based Moving Object Detection Network for Autonomous Driving
A novel two-stream architecture for joint learning of object detection and motion segmentation in a single convolutional network is proposed and it is shown that the joint training of tasks significantly improves accuracy compared to training them independently. Expand
Challenges in Designing Datasets and Validation for Autonomous Driving
This paper focuses on bad practices that often happen in the autonomous driving from an industrial deployment perspective and focuses on the datasets design and validation scheme for autonomous driving, where it would like to highlight the common problems, wrong assumptions, and steps towards avoiding them. Expand
Deep semantic segmentation for automated driving: Taxonomy, roadmap and challenges
A generic taxonomic survey of semantic segmentation algorithms and then discusses how it fits in the context of automated driving and the particular challenges of deploying it into a safety system which needs high level of accuracy and robustness are listed. Expand
Non-local Image Dehazing
This work proposes an algorithm, linear in the size of the image, deterministic and requires no training, that performs well on a wide variety of images and is competitive with other state-of-the-art methods on the single image dehazing problem. Expand