Lecture: Deep Convolutional Neural Networks
Stanford University06-Dec-2018Shubhang DesaiStanford Vision and Learning LabDeep CNNsLecture: Deep Convolutional Neural Networks1
Today’s agendaDeep convolutional networksHistory of CNNsCNN devArchitecture searchDeep CNNs 06-Dec-2018Stanford University2
Previously Input ImageFeatureExtractorPrediction 𝑦"Classifier𝑐)* ,Deep -Dec-2018𝑦Input LabelStanford University𝐶𝐸Loss Function𝐿Loss Value3
Previously Input ImageFeatureExtractorPrediction 𝑦"Classifier𝑐)* ,Deep -Dec-2018𝑦Input LabelStanford University𝐶𝐸Loss Function𝐿Loss Value4
Previously Input ImageFeatureExtractorPrediction 𝑦"Classifier𝑐)* ,Deep -Dec-2018𝑦Input LabelStanford University𝐶𝐸Loss Function𝐿Loss Value5
Previously Input ImageFeatureExtractorPrediction 𝑦"Classifier𝑐)* ,Deep -Dec-2018𝑦Input LabelStanford University𝐶𝐸Loss Function𝐿Loss Value3) Using gradient descent!6
Previously Why only one convolution?Input ImageFeatureExtractorPrediction 𝑦"Classifier𝑐)* ,Deep -Dec-2018𝑦Input LabelStanford University𝐶𝐸Loss Function𝐿Loss Value3) Using gradient descent!7
ConvolutionsConvolutions InsightsDeep CNNsMore Convolutions More Insights?06-Dec-2018Stanford University8
Recall Hubel and Weisel Deep CNNs06-Dec-2018Stanford University9
Recall Hubel and Weisel The thing hasedges It’s a mouse toy!Deep CNNsThe edges can be groupedinto triangles and ovals The triangles are ears, theoval is a body 06-Dec-2018Stanford University10
Recall Hubel and Weisel The thing hasedges It’s a mouse toy!Deep CNNsThe edges can be groupedinto triangles and ovals The triangles are ears, theoval is a body 06-Dec-2018Stanford University11
Convolutions Across ChannelsDeep CNNs06-Dec-201828 28 3 ImageStanford University15 15 3 Filter14 14 1 Output12
Convolutions Across ChannelsDeep CNNs06-Dec-201828 28 3 ImageStanford University15 15 3 4 Filter14 14 4 Output13
Convolutions Across ChannelsDeep CNNsmore output channels more filters more features we can learn!06-Dec-201828 28 3 ImageStanford University15 15 3 4 Filter14 14 4 Output14
Convolutions Across ChannelsDeep CNNs15 15 3 4 Conv Block06-Dec-2018Stanford University15
Stacking ConvolutionsStanford University28 28 4Output14 14 6Output7 7 8Output1 1 10Output06-Dec-201832 32 3Input7 7 8 10Conv Block8 8 6 8Conv BlockDeep CNNs15 15 4 6Conv Block5 5 3 4Conv Block16
Stacking ConvolutionsStanford University28 28 4Output14 14 6Output7 7 8Output1 1 10Output06-Dec-201832 32 3Input7 7 8 10Conv Block8 8 6 8Conv BlockDeep CNNs15 15 4 6Conv Block5 5 3 4Conv Block17
Convolutional Neural Networks (ConvNets) Often times end in fully-connected layers as the “classifier”Deep CNNs Neural networks which involve the stacking of multiple convolutional layers toproduce output06-Dec-2018Stanford University18
History of ConvNetsDeep CNNs06-Dec-2018LeNet – 1998Stanford University19
History of ConvNetsDeep CNNs06-Dec-2018AlexNet – 2012Stanford University20
History of ConvNetsDeep CNNs06-Dec-2018NiN – 2013Stanford University21
History of ConvNetsDeep CNNs06-Dec-2018Inception Network – 2015Stanford University22
Why Do They Work So Well?Deep CNNs06-Dec-2018Stanford University23
Why Do They Work So Well?Deep CNNs06-Dec-2018Stanford University24
Why Do They Work So Well?Deep CNNs06-Dec-2018Stanford University25
Why Do They Work So Well?Deep CNNs06-Dec-2018Stanford University26
Why Do They Work So Well?Deep CNNsThis is the neural network’s“receptive field”—it’s ableto see!06-Dec-2018Stanford University27
Great Applications of ConvNetsFine-Grained RecognitionDeep CNNs“StaffordshireBull Terrier”SegmentationStanford UniversityArt GenerationFacial Recognition06-Dec-2018“RanjayKrishna”28
What is CNN Dev? Define the objective Create the architecture– How many conv layers?– What size are the convolutions?– How many fully-connected layers?– What is the learning rate? Train and evaluate06-Dec-2018 Define hyperparametersDeep CNNs– What is the input/output?– What is the loss/objective function?– How did we do?– How can we do better?Stanford University29
What is CNN Dev? Define the objectiveDeep CNNs– What is the input/output?– What is the loss/objective function? Create the architecture– How many conv layers?– What size are the convolutions?– How many fully-connected layers?06-Dec-2018 Define hyperparameters– What is the learning rate? Train and evaluate– How did we do?– How can we do better?Stanford UniversityCan this be automated?30
Neural Architecture SearchAutomatically finds the best architecture for a given taskDeep CNNsBefore we had to find best featurizer for a fixed classifier—now we find the bestclassifier and featurizer in tandem!06-Dec-2018Stanford University31
In summary We can use convolutions as a basis to build powerful visual systemsDeep CNNsWe can leverage deep learning to automatically learn the best ways to do previouslydifficult tasks in computer visionStill lots of open questions!Stanford University06-Dec-2018If you’re interested in machine learning and/or deep learning, take: Machine Learning (CS 229) Deep Learning (CS 230) NLP with Deep Learning (CS 224n) Convolutional Neural Networks (CS 231n)32
Lecture: Deep Convolutional Neural Networks Shubhang Desai Stanford Vision and Learning Lab. s Stanford University 06-c-2018 2 Today’s agenda Deep convolutional networks . 28 28 3 Image 15 15 3 4 Filter 14 14 4 Output more output channels more filters more features we can learn! s Stanford University 06-c-
Learning a Deep Convolutional Network for Image Super-Resolution . a deep convolutional neural network (CNN) [15] that takes the low- . Convolutional Neural Networks. Convolutional neural networks (CNN) date back decades [15] and have recently shown an explosive popularity par-
Deep Neural Networks Convolutional Neural Networks (CNNs) Convolutional Neural Networks (CNN, ConvNet, DCN) CNN a multi‐layer neural network with – Local connectivity: Neurons in a layer are only connected to a small region of the layer before it – Share weight parameters across spatial positions:
Video Super-Resolution With Convolutional Neural Networks Armin Kappeler, Seunghwan Yoo, Qiqin Dai, and Aggelos K. Katsaggelos, Fellow, IEEE Abstract—Convolutional neural networks (CNN) are a special type of deep neural networks (DNN). They have so far been suc-cessfully applied to image super-resolution (SR) as well as other image .
Dual-domain Deep Convolutional Neural Networks for Image Demoireing An Gia Vien, Hyunkook Park, and Chul Lee Department of Multimedia Engineering Dongguk University, Seoul, Korea viengiaan@mme.dongguk.edu, hyunkook@mme.dongguk.edu, chullee@dongguk.edu Abstract We develop deep convolutional neural networks (CNNs)
Deep Convolutional Neural Networks for Remote Sensing Investigation of Looting of the Archeological Site of Al-Lisht, Egypt by Timberlynn Woolf . potential to expedite the looting detection process using Deep Convolutional Neural Networks (CNNs). Monitoring of looting is complicated in that it is an illicit activity, subject to legal sanction .
ImageNet Classification with Deep Convolutional Neural Networks, NIPS 2012 M. Zeiler and R. Fergus, Visualizing and Understanding Convolutional Networks, ECCV 2014 K. Simonyan and A. Zisserman, Very Deep Convolutional Networks for Large-Scale Image Recognition, ICLR 2015
2 Convolutional neural networks CNNs are hierarchical neural networks whose convolutional layers alternate with subsampling layers, reminiscent of sim-ple and complex cells in the primary visual cortex [Wiesel and Hubel, 1959]. CNNs vary in how convolutional and sub-sampling layers are realized and how the nets are trained. 2.1 Image processing .
API 526 provides effective discharge areas for a range of sizes in terms of letter designations, “D” through “T.” 3.19 Flutter Fluttering is where the PRV is open but the dynamics of the system cause abnormal, rapid reciprocating motion of the moveable parts of the PRV. During the fluttering, the disk does not contact the seat but reciprocates at the frequency of the flutter. 3.19 .
