Learning from Synthetic Data Using a Stacked Multichannel Autoencoder

Zhang, Xi; Fu, Yanwei; Jiang, Shanshan; Sigal, Leonid; Agam, Gady

doi:10.1109/icmla.2015.199

Cited by 35 publications

(37 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This behaviour could not be confirmed with our experiments. However, since our dataset is highly different from the dataset used in (Zhang et al, 2015), strong conclusions cannot be drawn.…”

Section: Testing Stepmentioning

confidence: 75%

“…Although standard procedures to prevent overfitting have been used, overfitting seems to be responsible for the unsatisfying performance. In the work of Zhang et al (Zhang, et al, 2015), the SVM classifier performed better in roof type classification compared to a classifier based on a CNN only. This behaviour could not be confirmed with our experiments.…”

Section: Testing Stepmentioning

confidence: 99%

“…Each patch is manually labelled related to the roof type. The main difference between this way of generating patches and previous methods (Zhang et al, 2015) and (Alidoost and Arefi, 2016) is that, the main direction of each roof is also considered inside the patch selection. In this way, the quality of roof patches cannot be degraded by rotation and resizing.…”

Section: Introductionmentioning

confidence: 99%

“…pixel intensities, texture, geometrical structures) to classify the building roofs into two categories: pitched and flat roofs. Zhang et al (2015) demonstrated that synthetic data could be used to improve the performance of the classifier. Roof type classification was performed based on recognizing detected edges of the roofs.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Roof Type Selection Based on Patch-Based Classification Using Deep Learning for High Resolution Satellite Imagery

Partovi

Fraundorfer

Azimi

et al. 2017

Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci.

View full text Add to dashboard Cite

ABSTRACT:3D building reconstruction from remote sensing image data from satellites is still an active research topic and very valuable for 3D city modelling. The roof model is the most important component to reconstruct the Level of Details 2 (LoD2) for a building in 3D modelling. While the general solution for roof modelling relies on the detailed cues (such as lines, corners and planes) extracted from a Digital Surface Model (DSM), the correct detection of the roof type and its modelling can fail due to low quality of the DSM generated by dense stereo matching. To reduce dependencies of roof modelling on DSMs, the pansharpened satellite images as a rich resource of information are used in addition. In this paper, two strategies are employed for roof type classification. In the first one, building roof types are classified in a state-of-the-art supervised pre-trained convolutional neural network (CNN) framework. In the second strategy, deep features from deep layers of different pre-trained CNN model are extracted and then an RBF kernel using SVM is employed to classify the building roof type. Based on roof complexity of the scene, a roof library including seven types of roofs is defined. A new semi-automatic method is proposed to generate training and test patches of each roof type in the library. Using the pre-trained CNN model does not only decrease the computation time for training significantly but also increases the classification accuracy.

show abstract

Section: Testing Stepmentioning

confidence: 75%

Section: Testing Stepmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Roof Type Selection Based on Patch-Based Classification Using Deep Learning for High Resolution Satellite Imagery

Partovi

Fraundorfer

Azimi

et al. 2017

Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci.

View full text Add to dashboard Cite

show abstract

“…They concluded that using such data for classifiers training is for the most part safe and will not result in worsening of classifier's accuracy. A practical method for learning image classifiers from synthetic data was presented by Zhang and others [13]. They identified a problem of distribution gap between real and synthetic data.…”

Section: Other Workmentioning

confidence: 99%

Evaluation of diagnostic classifiers using artificial clinical cases

2017

View full text Add to dashboard Cite

Abstract. Evaluation of classifiers in diagnosis support systems is a non-trivial task. It can be done in a form of controlled and blinded clinical trial, which is often difficult and costly. We propose a new method for generating artificial medical cases from a knowledge base, utilizing the concept of so-called medical diamonds. Cases generated using this method have features analogous to that of double-blinded trial and, thus, can be used for measuring sensitivity and specificity of diagnostic classifiers. This is easy and low-cost method of evaluation and comparison of classifiers in diagnosis support systems. We demonstrate that this method is able to produce valuable results when used for evaluation of similarity-based classifiers as well as shallow and deep neural networks.

show abstract

A Patch-Based Convolutional Neural Network for Localized MRI Brain Segmentation

Vambe

Viriri

Gwetu

2022

Lecture Notes of the Institute for Computer Sciences, Social Informatics and Telecommunications Engineering

View full text Add to dashboard Cite

The authors' contributions in the paper above are as follows:-1st Author : Concept formulation, literature review, design and implementation of algorithms and preparing and editing the write-up.-2nd and 3rd Authors : Concept improvement, providing advice, highlighting important considerations with respect to methods and techniques, and proof-reading manuscripts.I would like to express profound gratitude to my supervisors, that is, Prof. Serestina Viriri and Dr Mandlenkosi Gwetu, for their invaluable counsel, guidance and feedback during the work related to this dissertation. I would also like to extend heartfelt gratitude to several academics for their motivation and advice during my research. Last but not least I would like to thank my family for their continued support every step of the way. v 2.5D Two and a Half Dimensional. 5

show abstract

Learning from Synthetic Data Using a Stacked Multichannel Autoencoder

Cited by 35 publications

References 23 publications

Roof Type Selection Based on Patch-Based Classification Using Deep Learning for High Resolution Satellite Imagery

Roof Type Selection Based on Patch-Based Classification Using Deep Learning for High Resolution Satellite Imagery

Evaluation of diagnostic classifiers using artificial clinical cases

A Patch-Based Convolutional Neural Network for Localized MRI Brain Segmentation

Contact Info

Product

Resources

About