loading
Papers Papers/2022 Papers Papers/2022

Research.Publish.Connect.

Affiliation

Select a Subject

Forschungszentrum Jülich, Germany
Context (please use complete words only)
Field
Sort by

5 Papers

Rescuing Easy Samples in Self-Supervised Pretraining

Topic: Self-supervised Learning

Authors: Qin Wang 1 , Kai Krajsek 2 and Hanno Scharr 1

Affiliations: 1 IAS-8: Data Analytics and Machine Learning, Forschungszentrum Jülich, Germany , 2 Jülich Supercomputing Centre (JSC), Forschungszentrum Jülich, Germany

Digital Object Identifier: 10.5220/0013167900003912

Proceedings of VISAPP 2025, ISBN: 978-989-758-728-3

Collapse Button Quick Abstract
Many recent self-supervised pretraining methods use augmented versions of the same image as samples for their learning schemes. We observe that ’easy’ samples, i.e. samples being too similar to each other after augmentation, have only limited value as learning signal. We therefore propose to rescue easy samples and make them harder. To do so, we select the top k easiest samples using cosine similarity, strongly augment them, forward-pass them through the model, calculate cosine similarity of the output as loss, and add it to the original loss in a weighted fashion. This method can be adopted to all contrastive or other augmented-pair based learning methods, whether they involve negative pairs or not, as it changes handling of easy positives, only. This simple but effective approach introduces greater variability into such self-supervised pretraining processes, significantly increasing the performance on various downstream tasks as observed in our experiments. We pretrain models of different sizes, i.e. ResNet-50, ViT-S, ViT-B, or ViT-L, using ImageNet with SimCLR, MoCo v3, or DINOv2 training schemes. Here, e.g., we consistently find to improve results for ImageNet top-1 accuracy with a linear classifier establishing new SOTA for this task.

Likelihood Functions for Errors-in-variables Models - Bias-free Local Estimation with Minimum Variance

Topic: Optical Flow and Motion Analyses

Authors: Kai Krajsek , Christian Heinemann and Hanno Scharr

Affiliation: Forschungszentrum Jülich, Germany

Digital Object Identifier: 10.5220/0004667402700279

Proceedings of VISAPP 2014/VISIGRAPP, ISBN: 978-989-758-009-3

Collapse Button Quick Abstract
Parameter estimation in the presence of noisy measurements characterizes a wide range of computer vision problems. Thus, many of them can be formulated as errors-in-variables (EIV) problems. In this paper we provide a closed form likelihood function to EIV problems with arbitrary covariance structure. Previous approaches either do not offer a closed form, are restricted in the structure of the covariance matrix, or involve nuisance parameters. By using such a likelihood function, we provide a theoretical justification for well established estimators of EIV models. Furthermore we provide two maximum likelihood estimators for EIV parameters, a straight forward extension of a well known estimator and a novel, local estimator, as well as confidence bounds by means of the Cramer Rao Lower Bound. We show their performance by numerical experiments on optical flow estimation, as it is well explored and understood in literature. The straight forward extension turned out to have oscillating behavior, while the novel, local one performs favorably with respect to other methods. For small motions, it even performs better than an excellent global optical flow algorithm on the majority of pixel locations.
Collapse Button Related Ontology Subjects/Areas/Topics (3)
Computer Vision, Visualization and Computer Graphics

Motion, Tracking and Stereo Vision

Optical Flow and Motion Analyses

Using Channel Representations in Regularization Terms - A Case Study on Image Diffusion

Topic: Image Enhancement and Restoration

Authors: Christian Heinemann 1 , Freddie Åström 2 , George Baravdish 2 , Kai Krajsek 1 , Michael Felsberg 2 and Hanno Scharr 1

Affiliations: 1 Forschungszentrum Jülich, Germany , 2 Linköping University, Sweden

Digital Object Identifier: 10.5220/0004667500480055

Proceedings of VISAPP 2014/VISIGRAPP, ISBN: 978-989-758-003-1

Collapse Button Quick Abstract
In this work we propose a novel non-linear diffusion filtering approach for images based on their channel representation. To derive the diffusion update scheme we formulate a novel energy functional using a soft-histogram representation of image pixel neighborhoods obtained from the channel encoding. The resulting Euler-Lagrange equation yields a non-linear robust diffusion scheme with additional weighting terms stemming from the channel representation which steer the diffusion process. We apply this novel energy formulation to image reconstruction problems, showing good performance in the presence of mixtures of Gaussian and impulse-like noise, e.g. missing data. In denoising experiments of common scalar-valued images our approach performs competitive compared to other diffusion schemes as well as state-of-the-art denoising methods for the considered noise types.
Collapse Button Related Ontology Subjects/Areas/Topics (3)
Computer Vision, Visualization and Computer Graphics

Image Enhancement and Restoration

Image Formation and Preprocessing

3D RECONSTRUCTION OF PLANT ROOTS FROM MRI IMAGES

Topic: Medical Image Applications

Authors: Hannes Schulz 1 , Johannes A. Postma 2 , Dagmar van Dusschoten 2 , Hanno Scharr 2 and Sven Behnke 1

Affiliations: 1 University Bonn, Germany , 2 Forschungszentrum Jülich, Germany

Digital Object Identifier: 10.5220/0003869800240033

Proceedings of VISAPP 2012/VISIGRAPP, ISBN: 978-989-8565-04-4

Collapse Button Quick Abstract
We present a novel method for deriving a structural model of a plant root system from 3D Magnetic Resonance Imaging (MRI) data of soil grown plants. The structural model allows calculation of physiologically relevant parameters. Roughly speaking, MRI images show local water content of the investigated sample. The small, local amounts of water in roots require a relatively high resolution, which results in low SNR images. However, the spatial resolution of the MRI images remains coarse relative to the diameter of typical fine roots, causing many gaps in the visible root system. To reconstruct the root structure, we propose a three step approach: 1) detect tubular structures, 2) connect all pixels to the base of the root using Dijkstra’s algorithm, and 3) prune the tree using two signal strength related thresholds. Dijkstra’s algorithm determines the shortest path of each voxel to the base of the plant root, weighing the Euclidean distance measure by a multi-scale vesselness measure. As a result, paths running within good root candidates are preferred over paths in bare soil. We test this method using both virtually generated MRI images of Maize and real MRI images of Barley roots. In experiments on synthetic data, we show limitations of our algorithm with regard to resolution and noise levels. In addition we show how to use our reconstruction for root phenotyping on real MRI data of Barley roots in soil.
Collapse Button Related Ontology Subjects/Areas/Topics (3)
Applications and Services

Computer Vision, Visualization and Computer Graphics

Medical Image Applications

SUPPORTING SCIENTIFIC BIOLOGICAL APPLICATIONS WITH SEAMLESS DATABASE ACCESS IN INTEROPERABLE E-SCIENCE INFRASTRUCTURES

Topics: Databases and Data Management; Computational Molecular Biology; Algorithms and Software Tools

Authors: Sonja Holl , Morris Riedel , Bastian Demuth , Mathilde Romberg and Achim Streit

Affiliation: Forschungszentrum Juelich, Germany

Digital Object Identifier: 10.5220/0002695002030206

Proceedings of BIOINFORMATICS 2010/BIOSTEC, ISBN: 978-989-674-019-1

Collapse Button Quick Abstract
In the last decade, computational biological applications have become very well integrated into e-Science infrastructures. These distributed resources, containing computing and data sources, provide a reasonable environment for computing and data demanding applications. The access to e-Science infrastructures is mostly enstablished via Grids, where Grid clients support scientists using different types of resources. This paper extends an instance of the infrastructure interoperability reference model to remove the lack by adding centralized access to distributed computational and database resources via a graphical Grid client.
Collapse Button Related Ontology Subjects/Areas/Topics (5)
Algorithms and Software Tools

Biomedical Engineering

Databases and Data Management
Bioinformatics

Computational Molecular Systems

Topic(s):

Filter listed results, by

Publication date

From:
To:
Author
Locate Author in list below
Data pager
Data pager
12
 Page 1 of 2, items 1 to 10 of 16.
Add Author(s) to filter:

- Science and Technology Publications, Lda.
RESOURCES

Proceedings

Papers

Authors

Ontology

CONTACTS

Science and Technology Publications, Lda
Avenida de S. Francisco Xavier, Lote 7 Cv. C,
2900-616 Setúbal, Portugal.

Phone: +351 265 520 185 (National fixed network call)
Fax: +351 265 520 186
Email: info@scitepress.org

EXTERNAL LINKS

PRIMORIS

INSTICC

SCITEVENTS

CROSSREF

PROCEEDINGS SUBMITTED FOR INDEXATION BY:

dblp

Ei Compendex

SCOPUS

Semantic Scholar

Google Scholar

Microsoft Academic