PowerPoint PresentationTitle PowerPoint Presentation Author Jon Anning Created Date 11/17/2016...
1
Deep Discrete Flow Fatma Güney , Andreas Geiger , Autonomous Vision Group, MPI for Intelligent Systems, Tübingen Computer Vision and Geometry Group, ETH Zurich [email protected] Discrete Flow Feature Learning Using Dilated Convolutions Results and Remaining Problems M. Menze, C. Heipke, A. Geiger. Discrete Optimization for Optical Flow, GCPR 2015. J. Zbontar, Y. LeCun. Stereo Matching by Training a Convolutional Neural Network to Compare Image Patches, JMLR 2016. for i = 1to #dilations do Dilated Convolution with dilations[i] if i < #dilations then Batch Normalization, ReLU Sub-sampling with dilations[1] for i = 1 to #dilations do if i == #dilations then stride = 1 else stride = dilations[i+1] dilations[i] Convolution with stride if i < #dilations then Batch Normalization, ReLU Siamese Networks Convolution, ReLU Convolution, ReLU Convolution Normalization Second Patch Dot Product Convolution, ReLU Convolution, ReLU Convolution Normalization First Patch Similarity Score MPI Sintel KITTI
Transcript of PowerPoint PresentationTitle PowerPoint Presentation Author Jon Anning Created Date 11/17/2016...
-
Deep Discrete FlowFatma Güney
𝟏, Andreas Geiger
𝟏,𝟐
𝟏Autonomous Vision Group, MPI for Intelligent Systems, Tübingen
𝟐Computer Vision and Geometry Group, ETH Zurich
Discrete Flow
Feature Learning Using Dilated Convolutions
Results and Remaining Problems
M. Menze, C. Heipke, A. Geiger. Discrete Optimization for Optical Flow, GCPR 2015.
J. Zbontar, Y. LeCun. Stereo Matching by Training a Convolutional Neural Network to Compare Image Patches, JMLR 2016.
for i = 1 to #dilations doDilated Convolution with dilations[i]if i < #dilations then
Batch Normalization, ReLU
Sub-sampling with dilations[1]for i = 1 to #dilations do
if i == #dilations thenstride = 1
else
stride = dilations[i+1]dilations[i]
Convolution with strideif i < #dilations then
Batch Normalization, ReLU
Siamese Networks
Convolution, ReLU
Convolution, ReLU
Convolution
Normalization
Second Patch
Dot Product
Convolution, ReLU
Convolution, ReLU
Convolution
Normalization
First Patch
Similarity Score
MPI Sintel
KITTI
