Hochschulschrift
Convolutional networks to relate images
Zusammenfassung: Within recent years, convolutional neural networks have experienced a come back and are currently dominating the field of computer vision.In contrast to previously used methods, they allow for full data-based end-to-end training without the need to manually engineer features.On the downside, huge amounts of annotated data are needed to train a network successfully.Furthermore, the architectural design of these networks and other hyperparameters are sometimes hard to choose, especially when predicting high-resolution outputs from images.In this thesis, we show how networks can be trained without manually labeled data by using surrogate tasks, strong augmentation and synthetically generated training data.We demonstrate how generic image features for classification and matching can be learned unsupervisedly by augmenting random images to form surrogate classes.We also train networks to estimate the canonical orientation of rotated images by augmenting a large set of natural unlabeled images.In the main part of this thesis, we present a new network architecture that can be used for any pixel level regression task.We prove its value by creating the first end-to-end learning approach to compute optical flow and extend it to disparity and scene flow.All of this is done by using synthetically generated training data which we rendered using a custom version of the Blender rendering suite
- Location
-
Deutsche Nationalbibliothek Frankfurt am Main
- Extent
-
Online-Ressource
- Language
-
Englisch
- Notes
-
Albert-Ludwigs-Universität Freiburg, Dissertation, 2016
- Classification
-
Informatik
- Keyword
-
Optischer Fluss
- Event
-
Veröffentlichung
- (where)
-
Freiburg
- (who)
-
Universität
- (when)
-
2016
- Creator
- Contributor
- DOI
-
10.6094/UNIFR/11418
- URN
-
urn:nbn:de:bsz:25-freidok-114183
- Rights
-
Der Zugriff auf das Objekt ist unbeschränkt möglich.
- Last update
-
25.03.2025, 1:57 PM CET
Data provider
Deutsche Nationalbibliothek. If you have any questions about the object, please contact the data provider.
Object type
- Hochschulschrift
Associated
- Fischer, Philipp
- Brox, Thomas
- Universität
Time of origin
- 2016
Other Objects (12)
Estimating optical flow with convolutional neural networks
Biopolymer segmentation from CLSM microscopy images using a convolutional neural network
Convolutional Neural Networks for Classification and Segmentation of Medical Images
Fully Convolutional Networks for Semantic Segmentation from RGB-D images
Automatic electroencephalogram artifact removal using deep convolutional neural networks
Encoding candlesticks as images for pattern classification using convolutional neural networks
Applied machine learning for liver surgery : : The prediction of liver function from routine CT-images with convolutional neural networks
Interpretable functional specialization emerges in deep convolutional networks trained on brain signals
Assessing the semantic similarity of images of silk fabrics using convolutional neural networks
Supervised detection of bomb craters in historical aerial images using convolutional neural networks
Fast automated detection of COVID-19 from medical images using convolutional neural networks
Shared enhancer gene regulatory networks between wound and oncogenic programs
Estimating optical flow with convolutional neural networks
Biopolymer segmentation from CLSM microscopy images using a convolutional neural network
Convolutional Neural Networks for Classification and Segmentation of Medical Images
Fully Convolutional Networks for Semantic Segmentation from RGB-D images
Automatic electroencephalogram artifact removal using deep convolutional neural networks
Encoding candlesticks as images for pattern classification using convolutional neural networks
Applied machine learning for liver surgery : : The prediction of liver function from routine CT-images with convolutional neural networks
Interpretable functional specialization emerges in deep convolutional networks trained on brain signals
Assessing the semantic similarity of images of silk fabrics using convolutional neural networks
Supervised detection of bomb craters in historical aerial images using convolutional neural networks
Fast automated detection of COVID-19 from medical images using convolutional neural networks
Shared enhancer gene regulatory networks between wound and oncogenic programs
Estimating optical flow with convolutional neural networks
Biopolymer segmentation from CLSM microscopy images using a convolutional neural network
Convolutional Neural Networks for Classification and Segmentation of Medical Images
Fully Convolutional Networks for Semantic Segmentation from RGB-D images
Automatic electroencephalogram artifact removal using deep convolutional neural networks
Encoding candlesticks as images for pattern classification using convolutional neural networks
Applied machine learning for liver surgery : : The prediction of liver function from routine CT-images with convolutional neural networks
Interpretable functional specialization emerges in deep convolutional networks trained on brain signals
Assessing the semantic similarity of images of silk fabrics using convolutional neural networks
Supervised detection of bomb craters in historical aerial images using convolutional neural networks
Fast automated detection of COVID-19 from medical images using convolutional neural networks