Robustness of Deep Convolutional Neural Networks for Image Degradations

Ghosh, Sanjukta; Shet, Rohan; Amon, Peter; Hutter, Andreas; Kaup, André

doi:10.1109/icassp.2018.8461907

Cited by 45 publications

(23 citation statements)

References 6 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Remote sensing image data with more than three bands have not yet been able to be trained in deep learning training networks. Specifically, deep learning (DL)-based past researchers are designed to accept standard RGB bands as they confront with everyday images [ 73 , 74 , 75 ]. Moreover, in terms of using multispectral perspective, the Arctic tundra vegetation communities have separable view in Arctic mapping application [ 76 , 77 ].…”

Section: Model Evaluation Results and Discussionmentioning

confidence: 99%

Use of Very High Spatial Resolution Commercial Satellite Imagery and Deep Learning to Automatically Map Ice-Wedge Polygons across Tundra Vegetation Types

2020

View full text Add to dashboard Cite

We developed a high-throughput mapping workflow, which centers on deep learning (DL) convolutional neural network (CNN) algorithms on high-performance distributed computing resources, to automatically characterize ice-wedge polygons (IWPs) from sub-meter resolution commercial satellite imagery. We applied a region-based CNN object instance segmentation algorithm, namely the Mask R-CNN, to automatically detect and classify IWPs in North Slope of Alaska. The central goal of our study was to systematically expound the DLCNN model interoperability across varying tundra types (sedge, tussock sedge, and non-tussock sedge) and image scene complexities to refine the understanding of opportunities and challenges for regional-scale mapping applications. We corroborated quantitative error statistics along with detailed visual inspections to gauge the IWP detection accuracies. We found promising model performances (detection accuracies: 89% to 96% and classification accuracies: 94% to 97%) for all candidate image scenes with varying tundra types. The mapping workflow discerned the IWPs by exhibiting low absolute mean relative error (AMRE) values (0.17–0.23). Results further suggest the importance of increasing the variability of training samples when practicing transfer-learning strategy to map IWPs across heterogeneous tundra cover types. Overall, our findings demonstrate the robust performances of IWPs mapping workflow in multiple tundra landscapes.

show abstract

Section: Model Evaluation Results and Discussionmentioning

confidence: 99%

Use of Very High Spatial Resolution Commercial Satellite Imagery and Deep Learning to Automatically Map Ice-Wedge Polygons across Tundra Vegetation Types

2020

View full text Add to dashboard Cite

show abstract

“…Due to the mobility of vehicles, the images captured by on-board cameras generally suffer from motion blur, noise, and distortion [21], [22]. The noise and distortion in different vehicular clients may follow identical statistical distribution, while the motion blur level varies with instantaneous velocity of each vehicular client [7].…”

Section: A Image Qualitymentioning

confidence: 99%

Federated Learning in Vehicular Edge Computing: A Selective Model Aggregation Approach

et al. 2020

View full text Add to dashboard Cite

Federated learning is a newly emerged distributed machine learning paradigm, where the clients are allowed to individually train local deep neural network (DNN) models with local data and then jointly aggregate a global DNN model at the central server. Vehicular edge computing (VEC) aims at exploiting the computation and communication resources at the edge of vehicular networks. Federated learning in VEC is promising to meet the ever-increasing demands of artificial intelligence (AI) applications in intelligent connected vehicles (ICV). Considering image classification as a typical AI application in VEC, the diversity of image quality and computation capability in vehicular clients potentially affects the accuracy and efficiency of federated learning. Accordingly, we propose a selective model aggregation approach, where ''fine'' local DNN models are selected and sent to the central server by evaluating the local image quality and computation capability. Regarding the implementation of model selection, the central server is not aware of the image quality and computation capability in the vehicular clients, whose privacy is protected under such a federated learning framework. To overcome this information asymmetry, we employ two-dimension contract theory as a distributed framework to facilitate the interactions between the central server and vehicular clients. The formulated problem is then transformed into a tractable problem through successively relaxing and simplifying the constraints, and eventually solved by a greedy algorithm. Using two datasets, i.e., MNIST and BelgiumTSC, our selective model aggregation approach is demonstrated to outperform the original federated averaging (FedAvg) approach in terms of accuracy and efficiency. Meanwhile, our approach also achieves higher utility at the central server compared with the baseline approaches. INDEX TERMS Federated learning, vehicular edge computing, model aggregation, contract theory.

show abstract

“…In [12], the convolution layers of a classification network that performed worse on distorted input data according to a self-defined metric were retrained to improve the detection accuracy for distorted images. Another approach in [13] proposed adapting weights according to the incoming image degradation by a master-slave architecture.…”

Section: Decision Algorithmmentioning

confidence: 99%

Robust Deep Neural Object Detection and Segmentation for Automotive Driving Scenario with Compressed Image Data

Fischer

Blum

Herglotz

et al. 2021

2021 IEEE International Symposium on Circuits and Systems (ISCAS)

Self Cite

View full text Add to dashboard Cite

Deep neural object detection or segmentation networks are commonly trained with pristine, uncompressed data. However, in practical applications the input images are usually deteriorated by compression that is applied to efficiently transmit the data. Thus, we propose to add deteriorated images to the training process in order to increase the robustness of the two state-of-the-art networks Faster and Mask R-CNN. Throughout our paper, we investigate an autonomous driving scenario by evaluating the newly trained models on the Cityscapes dataset that has been compressed with the upcoming video coding standard Versatile Video Coding (VVC). When employing the models that have been trained with the proposed method, the weighted average precision of the R-CNNs can be increased by up to 3.68 percentage points for compressed input images, which corresponds to bitrate savings of nearly 48 %.

show abstract

Robustness of Deep Convolutional Neural Networks for Image Degradations

Cited by 45 publications

References 6 publications

Use of Very High Spatial Resolution Commercial Satellite Imagery and Deep Learning to Automatically Map Ice-Wedge Polygons across Tundra Vegetation Types

Use of Very High Spatial Resolution Commercial Satellite Imagery and Deep Learning to Automatically Map Ice-Wedge Polygons across Tundra Vegetation Types

Federated Learning in Vehicular Edge Computing: A Selective Model Aggregation Approach

Robust Deep Neural Object Detection and Segmentation for Automotive Driving Scenario with Compressed Image Data

Contact Info

Product

Resources

About