5,376 Hits in 3.8 sec

Calibrating Deep Neural Networks using Focal Loss [article]

Jishnu Mukhoti, Viveka Kulharia, Amartya Sanyal, Stuart Golodetz, Philip H.S. Torr, Puneet K. Dokania
2020 arXiv   pre-print
Miscalibration - a mismatch between a model's confidence and its correctness - of Deep Neural Networks (DNNs) makes their predictions hard to rely on.  ...  We show that, as opposed to the standard cross-entropy loss, focal loss [Lin et. al., 2017] allows us to learn models that are already very well calibrated.  ...  Appendix: Calibrating Deep Neural Networks using Focal Loss In §A, we provide some empirical evidence for the observation made in §3 in the main paper using reliability plots.  ... 
arXiv:2002.09437v2 fatcat:rp3zep7fyren5j6jcyuanwwfe4

A Calibrated Multiexit Neural Network for Detecting Urothelial Cancer Cells

L. Lilli, E. Giarnieri, S. Scardapane, Nadia A. Chuzhanova
2021 Computational and Mathematical Methods in Medicine  
calibrated than a baseline deep convolutional network.  ...  Deep convolutional networks have become a powerful tool for medical imaging diagnostic.  ...  Acknowledgments This work was supported in part by the project Detection of urothelial cancer cells using deep convolutional neural networks (Progetti di Ricerca Sapienza).  ... 
doi:10.1155/2021/5569458 pmid:34234839 pmcid:PMC8216797 fatcat:algargklsrabheapnckphabwmi

Dynamically Weighted Balanced Loss: Class Imbalanced Learning and Confidence Calibration of Deep Neural Networks

K. Ruwani M. Fernando, Chris P. Tsokos
2021 IEEE Transactions on Neural Networks and Learning Systems  
We further show that the proposed loss function is classification calibrated.  ...  To address the class distribution imbalance in deep learning, we propose a class rebalancing strategy based on a class-balanced dynamically weighted loss function where weights are assigned based on the  ...  DEEP NEURAL NETWORK PRELIMINARIES Through several underlying network blocks or layers, Deep Neural Networks (DNNs) extract representative features and hidden structural knowledge from data automatically  ... 
doi:10.1109/tnnls.2020.3047335 pmid:33444149 fatcat:c3y7y4p4afdppbo5pm4dgbjhpi

Maximum Entropy on Erroneous Predictions (MEEP): Improving model calibration for medical image segmentation [article]

Agostina Larrazabal, Cesar Martinez, Jose Dolz, Enzo Ferrante
2021 arXiv   pre-print
Modern deep neural networks have achieved remarkable progress in medical image segmentation tasks.  ...  Our method is agnostic to the neural architecture, does not increase model complexity and can be coupled with multiple segmentation loss functions.  ...  Furthermore, we also compare our method to stateof-the-art approaches that have proven to provide better calibrated deep neural networks.  ... 
arXiv:2112.12218v1 fatcat:kegmdn6obrb5rfx3jyrhva3vka

Camera Calibration through Camera Projection Loss [article]

Talha Hanif Butt, Murtaza Taj
2022 arXiv   pre-print
We estimate the desired parameters via novel camera projection loss (CPL) that uses the camera model neural network to reconstruct the 3D points and uses the reconstruction loss to estimate the camera  ...  Our code and generated dataset are available at  ...  We train a convolutional neural network to predict the extrinsic and intrinsic camera parameters.  ... 
arXiv:2110.03479v3 fatcat:gbwqjow2yzfcdcytwwn7bvinra

Single Image Automatic Radial Distortion Compensation Using Deep Convolutional Network [article]

Igor Janos, Wanda Benesova
2021 arXiv   pre-print
Keywords: Deep Convolutional Neural Network, Radial Distortion, Single Image Rectification  ...  We present a novel method for single-image automatic lens distortion compensation based on deep convolutional neural networks, capable of real-time performance and accuracy using two highest-order coefficients  ...  We train our deep neural network to directly estimate the values of k2 . Network architecture.  ... 
arXiv:2112.08198v1 fatcat:df6o4xbstncgvce2ineglxwo7i

DEEPFOCAL: A method for direct focal length estimation

Scott Workman, Connor Greenwell, Menghua Zhai, Ryan Baltenberger, Nathan Jacobs
2015 2015 IEEE International Conference on Image Processing (ICIP)  
In this work, we explore the application of a deep convolutional neural network, trained on natural images obtained from Internet photo collections, to directly estimate the focal length using only raw  ...  or a calibration grid, to occur in the field of view.  ...  We proposed a fast method which overcomes these limitations by directly estimating the focal length from raw pixels using a deep convolutional neural network.  ... 
doi:10.1109/icip.2015.7351024 dblp:conf/icip/WorkmanGZBJ15 fatcat:ygjq3dwtz5g3hpd4jxijdavtbu

Soft Calibration Objectives for Neural Networks [article]

Archit Karandikar, Nicholas Cain, Dustin Tran, Balaji Lakshminarayanan, Jonathon Shlens, Michael C. Mozer, Becca Roelofs
2021 arXiv   pre-print
However, deep neural networks are often under- or over-confident in their predictions.  ...  Overall, experiments across losses and datasets demonstrate that using calibration-sensitive procedures yield better uncertainty estimates under dataset shift than the standard practice of using a cross  ...  Acknowledgements The authors thank Brennan McConnell and Mohammad Khajah who conducted initial explorations of soft binning calibration loss.  ... 
arXiv:2108.00106v2 fatcat:oxgu3qitdzb55b7lehp5ces5ii

A comprehensive study on the prediction reliability of graph neural networks for virtual screening [article]

Soojung Yang, Kyung Hoon Lee, Seongok Ryu
2020 arXiv   pre-print
Prediction models based on deep neural networks are increasingly gaining attention for fast and accurate virtual screening systems.  ...  We investigate the effects of model architectures, regularization methods, and loss functions on the prediction performance and reliability of classification results.  ...  for deep neural networks.  ... 
arXiv:2003.07611v1 fatcat:ntyjvwu4dzfvzemixhuyaotmpm

Identifying and Exploiting Structures for Reliable Deep Learning [article]

Amartya Sanyal
2021 arXiv   pre-print
To do this, we identify structures in deep neural networks that can be exploited to mitigate the above causes of unreliability of deep learning algorithms.  ...  The extraordinary performance of these systems often gives the impression that they can be used to revolutionise our lives for the better.  ...  Approaches for calibrating deep neural networks Multiple approaches have been proposed in the literature for calibrating a neural network in practice.  ... 
arXiv:2108.07083v1 fatcat:lducrn5tlfeqvpxevz6gukfvse

Assignment of Focus Position with Convolutional Neural Networks in Adaptive Lens Based Axial Scanning for Confocal Microscopy

Katharina Schmidt, Nektarios Koukourakis, Jürgen W. Czarske
2022 Applied Sciences  
However, maybe the training procedure of the neural network must be adapted for some use cases.  ...  Here, we introduce an alternative approach which provides a single shot estimation of the current axial focus position by a convolutional neural network.  ...  Deep Single Image Camera Calibration with Radial Distortion.  ... 
doi:10.3390/app12020661 fatcat:yptctmp5krgb7oz6abwcy2xroa

SqueezeSegV2: Improved Model Structure and Unsupervised Domain Adaptation for Road-Object Segmentation from a LiDAR Point Cloud [article]

Bichen Wu, Xuanyu Zhou, Sicheng Zhao, Xiangyu Yue, Kurt Keutzer
2018 arXiv   pre-print
Earlier work demonstrates the promise of deep-learning-based approaches for point cloud segmentation; however, these approaches need to be improved to be practically useful.  ...  With improved model structure, training loss, batch normalization and additional input channel, SqueezeSegV2 achieves significant accuracy improvement when trained on real data.  ...  ACKNOWLEDGEMENT This work is partially supported by Berkeley Deep Drive (BDD), and partially sponsored by individual gifts from Intel and Samsung.  ... 
arXiv:1809.08495v1 fatcat:jse7z2stzbgr5exvy5k3qw4vxi

High Precision Calibration Algorithm for Binocular Stereo Vision Camera using Deep Reinforcement Learning

Jie Ren, Fuyu Guan, Tingting Wang, Baoshan Qian, Chunlin Luo, Guoliang Cai, Ce Kan, Xiaofeng Li, Zaher Mundher Yaseen
2022 Computational Intelligence and Neuroscience  
The deep learning fitting value function is used based on the internal parameters.  ...  To address the issue of insufficient precision, therefore, a high precision calibration algorithm for binocular stereo vision camera using deep reinforcement learning is proposed.  ...  neural networks.  ... 
doi:10.1155/2022/6596868 pmid:35401726 pmcid:PMC8989564 fatcat:fboyxxthdbffrklfq4td4oiy5a

AnatomyNet: Deep learning for fast and fully automated whole-volume segmentation of head and neck anatomy

Wentao Zhu, Yufang Huang, Liang Zeng, Xuming Chen, Yong Liu, Zhen Qian, Nan Du, Wei Fan, Xiaohui Xie
2018 Medical Physics (Lancaster)  
function combining Dice scores and focal loss to facilitate the training of the neural model.  ...  Results: We collected 261 HaN CT images to train AnatomyNet, and used MICCAI Head and Neck Auto Segmentation Challenge 2015 as a benchmark dataset to evaluate the performance of AnatomyNet.  ...  Ibragimov and Xing proposed a simple convolutional neural network for atlas-free deep learning based OARs segmentation [26] .  ... 
doi:10.1002/mp.13300 fatcat:n7nqws6jdjbqrfkqdjzpu3njde

Effect Of Personalized Calibration On Gaze Estimation Using Deep-Learning [article]

Nairit Bandyopadhyay, Sébastien Riou, Didier Schwab
2021 arXiv   pre-print
We trained a multi modal convolutional neural network and analysed its performance with and without calibration and this evaluation provides clear insights on how calibration improved the performance of  ...  To analyse the performance in such scenarios we have tried to simulate a calibration mechanism. In this work we use the MPIIGaze data set.  ...  Thus with the neural network we are solving an optimisation problem of minimising this loss function The normalized data set is available online for public use.  ... 
arXiv:2109.12801v1 fatcat:dirot2hdirf3thgxavglxrulgq
« Previous Showing results 1 — 15 out of 5,376 results