ai-driven analysis and image acquisition of in-vivo ...€¦ · chen, krämer, klein, romen –...
TRANSCRIPT
Department of MathematicsTechnical University of Munich
AI-driven analysis and image acquisition of in-vivoneuronal network activity
Xingying Chen, Bastian KrämerMichal Klein, René Romen
August 5, 2019
Department of MathematicsTechnical University of Munich
Department of MathematicsTechnical University of Munich
Table of Contents
1 Introduction
2 Data
3 Networks
4 Results
5 Discussion
Chen, Krämer, Klein, Romen – Data Innovation Lab 2
Department of MathematicsTechnical University of Munich
Table of Contents
1 Introduction
2 Data
3 Networks
4 Results
5 Discussion
Chen, Krämer, Klein, Romen – Data Innovation Lab 3
Department of MathematicsTechnical University of Munich
Zebrafish Larvae
Figure 1: Zebrafish[5]
Chen, Krämer, Klein, Romen – Data Innovation Lab 4
Department of MathematicsTechnical University of Munich
Light Field Microscope
Figure 2: (a) A microscopewith a microlens array[3]
Figure 3: Microlens array[1]
Chen, Krämer, Klein, Romen – Data Innovation Lab 5
Department of MathematicsTechnical University of Munich
Advantages of Light Field Microscopy Images
Figure 4: Comparison of images at di�erent focal planes[3]
Chen, Krämer, Klein, Romen – Data Innovation Lab 6
Department of MathematicsTechnical University of Munich
Light Field Microscope
Figure 5: (b), (c) The red point stands for a point source generating illumination. Redregions on top shows the intensity of illumination arriving at the sensor plane[3].
Chen, Krämer, Klein, Romen – Data Innovation Lab 7
Department of MathematicsTechnical University of Munich
Light Field Microscope
Figure 6: Example light field of a sphere[3]
Chen, Krämer, Klein, Romen – Data Innovation Lab 8
Department of MathematicsTechnical University of Munich
Light Field Transformations
43
Lenslet Image
43
Lenslet Sub-image
23
23I Angular resolution: 43× 43
I Spatial resolution: 23× 23
Chen, Krämer, Klein, Romen – Data Innovation Lab 9
Department of MathematicsTechnical University of Munich
Reordering Light Field
Figure 7: Reordering lenslet to views image[6]
Chen, Krämer, Klein, Romen – Data Innovation Lab 10
Department of MathematicsTechnical University of Munich
Reordering Light Field
Figure 8: Reordering lenslet to views image[6]
Chen, Krämer, Klein, Romen – Data Innovation Lab 11
Department of MathematicsTechnical University of Munich
Reordering Light Field
Figure 9: Reordering lenslet to views image[6]
Chen, Krämer, Klein, Romen – Data Innovation Lab 12
Department of MathematicsTechnical University of Munich
Light Field Transformations
43
View
43
Views Image
23
23I Angular resolution: 43× 43
I Spatial resolution: 23× 23
Chen, Krämer, Klein, Romen – Data Innovation Lab 13
Department of MathematicsTechnical University of Munich
Light Field Image
43
Lenslet Image
43
Views Image
23
23
Views
Chen, Krämer, Klein, Romen – Data Innovation Lab 14
Department of MathematicsTechnical University of Munich
Goal
Chen, Krämer, Klein, Romen – Data Innovation Lab 15
Department of MathematicsTechnical University of Munich
Goal
Chen, Krämer, Klein, Romen – Data Innovation Lab 16
Department of MathematicsTechnical University of Munich
Goal
Chen, Krämer, Klein, Romen – Data Innovation Lab 17
Department of MathematicsTechnical University of Munich
Goal
Chen, Krämer, Klein, Romen – Data Innovation Lab 18
Department of MathematicsTechnical University of Munich
Table of Contents
1 Introduction
2 Data
3 Networks
4 Results
5 Discussion
Chen, Krämer, Klein, Romen – Data Innovation Lab 19
Department of MathematicsTechnical University of Munich
Datasets - Real Images
I Fish eyeLenslet Views
Figure 10: Fish eye
Chen, Krämer, Klein, Romen – Data Innovation Lab 20
Department of MathematicsTechnical University of Munich
Datasets - Real Images
I SpheresLenslet Views
Figure 11: SpheresChen, Krämer, Klein, Romen – Data Innovation Lab 21
Department of MathematicsTechnical University of Munich
Dataset - Training Data
I Created own dataset containing geometric objects
Figure 12: Scene
Ground Truth Depth Map
Figure 13: Resulting depth map
Chen, Krämer, Klein, Romen – Data Innovation Lab 22
Department of MathematicsTechnical University of Munich
Dataset
Figure 14: Resulting lenslet and views image
I 369 images
Chen, Krämer, Klein, Romen – Data Innovation Lab 23
Department of MathematicsTechnical University of Munich
Table of Contents
1 Introduction
2 Data
3 Networks
4 Results
5 Discussion
Chen, Krämer, Klein, Romen – Data Innovation Lab 24
Department of MathematicsTechnical University of Munich
Overview of Networks
I Input - views imageI Epinet
I Views network
I Input - lenslet imageI Lenslet network
I Lenslet classification network
Chen, Krämer, Klein, Romen – Data Innovation Lab 25
Department of MathematicsTechnical University of Munich
Overview of Networks
I Input - views imageI Epinet
I Views network
I Input - lenslet imageI Lenslet network
I Lenslet classification network
Chen, Krämer, Klein, Romen – Data Innovation Lab 26
Department of MathematicsTechnical University of Munich
Epinet
Figure 15: Epinet architecture [9]
I Multi-stream network
Chen, Krämer, Klein, Romen – Data Innovation Lab 27
Department of MathematicsTechnical University of Munich
Epinet
I 2.2 million parameters
I Multistream convolutional encoder
I Upscaling: nearest neighbor interpolation
I Optimizer: RMSProp (no learning rate decay)
I Regularization: None
I Activation function: Leaky ReLU
I Learning rate: 10−4
I Error: Mean squared error
Chen, Krämer, Klein, Romen – Data Innovation Lab 28
Department of MathematicsTechnical University of Munich
Overview of Networks
I Input - views imageI Epinet
I Views network
I Input - lenslet imageI Lenslet network
I Lenslet classification network
Chen, Krämer, Klein, Romen – Data Innovation Lab 29
Department of MathematicsTechnical University of Munich
Views Network
52943
43
256
1
1
256
2
2
256
4
4
128
4
4
128
256
256
64
16
16
32
4
4
1
1
1
Figure 16: Le� to right: input, 4 convolution layers, upscaling by nearest neighborinterpolation, 3 convolution layers.
Chen, Krämer, Klein, Romen – Data Innovation Lab 30
Department of MathematicsTechnical University of Munich
Views Network
I 4.1 million parameters
I Convolutional encoder
I Upscaling: nearest neighbor interpolation
I Optimizer: Adam (no learning rate decay)
I Regularization: None
I Activation function: Leaky ReLU
I Learning rate: 10−6
I Error: Absolute error
Chen, Krämer, Klein, Romen – Data Innovation Lab 31
Department of MathematicsTechnical University of Munich
Overview of Networks
I Input - views imageI Epinet
I Views network
I Input - lenslet imageI Lenslet network
I Lenslet classification network
Chen, Krämer, Klein, Romen – Data Innovation Lab 32
Department of MathematicsTechnical University of Munich
Lenslet Network
Figure 17: General structure of a convolution encoder-decoder architecture. We useInception-ResNet-v2 [10] blocks in encoder.
Chen, Krämer, Klein, Romen – Data Innovation Lab 33
Department of MathematicsTechnical University of Munich
Lenslet Network
I 7.6 million parameters
I Inception-ResNet-v2 blocks encoder
I Upscaling: fast up-convolution (x3)
I Optimizer: Adam (exp. decaying learning rate (400 examples))
I Regularization: L2 (scale: 5 ∗ 10−6)
I Activation function: Leaky ReLU (a�er each block)
I Learning rate: 5 ∗ 10−6
I Error: Mean squared error
Chen, Krämer, Klein, Romen – Data Innovation Lab 34
Department of MathematicsTechnical University of Munich
Overview of Networks
I Input - views imageI Epinet
I Views network
I Input - lenslet imageI Lenslet network
I Lenslet classification network
Chen, Krämer, Klein, Romen – Data Innovation Lab 35
Department of MathematicsTechnical University of Munich
Lenslet Classification Network
Figure 18: General structure of a convolution encoder-decoder architecture. We useInception-ResNet-v2[10] blocks in encoder.
Chen, Krämer, Klein, Romen – Data Innovation Lab 36
Department of MathematicsTechnical University of Munich
Lenslet Classification Network
I 5 million parameters
I Inception-ResNet-v2 encoder
I Upscaling: fast up-convolution (x3)
I Optimizer: Adam (exp. decaying learning rate (400 examples))
I Regularization: L2 (scale: 5 ∗ 10−6)
I Activation function: Leaky ReLU (a�er each block)
I Learning rate: 5 ∗ 10−6
I Error: Mean absolute error
Chen, Krämer, Klein, Romen – Data Innovation Lab 37
Department of MathematicsTechnical University of Munich
Table of Contents
1 Introduction
2 Data
3 Networks
4 Results
5 Discussion
Chen, Krämer, Klein, Romen – Data Innovation Lab 38
Department of MathematicsTechnical University of Munich
�antitative Results
Network MSE BPR (δ = 0.15) Input #params (millions)Epinet 0.0049958 0.04304 Stacked views 2.2Views 0.0122427 0.12022 Stacked views 4.1Lenslet 0.00595414 0.05586 Lenslet 7.6
Lenslet (cls) 26.9925 0.99521 Lenslet 5
Chen, Krämer, Klein, Romen – Data Innovation Lab 39
Department of MathematicsTechnical University of Munich
�alitative Results - Test DataGround Truth Epinet Views
Lenslet Lenslet (cls)
0.0
0.2
0.4
0.6
0.8
1.0
Figure 19: Depth prediction of an image from test set a�er 300 epochs.
Chen, Krämer, Klein, Romen – Data Innovation Lab 40
Department of MathematicsTechnical University of Munich
�alitative Results - Fish EyeEpinet Views
Lenslet Lenslet (cls)
0.0
0.2
0.4
0.6
0.8
1.0
Figure 20: Depth prediction of fish eye a�er 300 epochs.
Chen, Krämer, Klein, Romen – Data Innovation Lab 41
Department of MathematicsTechnical University of Munich
�alitative Results - SpheresEpinet Views
Lenslet Lenslet (cls)
0.0
0.2
0.4
0.6
0.8
1.0
Figure 21: Depth prediction of spheres a�er 300 epochs.
Chen, Krämer, Klein, Romen – Data Innovation Lab 42
Department of MathematicsTechnical University of Munich
Discussion - Real Data
I Image of fish eye and spheres taken with di�erent configurations
I Be�er networks on the test set pick up more noise
I No ground truth to compare the results
Chen, Krämer, Klein, Romen – Data Innovation Lab 43
Department of MathematicsTechnical University of Munich
Discussion - Real Data
I Image of fish eye and spheres taken with di�erent configurations
I Be�er networks on the test set pick up more noise
I No ground truth to compare the results
Chen, Krämer, Klein, Romen – Data Innovation Lab 43
Department of MathematicsTechnical University of Munich
Discussion - Real Data
I Image of fish eye and spheres taken with di�erent configurations
I Be�er networks on the test set pick up more noise
I No ground truth to compare the results
Chen, Krämer, Klein, Romen – Data Innovation Lab 43
Department of MathematicsTechnical University of Munich
Table of Contents
1 Introduction
2 Data
3 Networks
4 Results
5 Discussion
Chen, Krämer, Klein, Romen – Data Innovation Lab 44
Department of MathematicsTechnical University of Munich
Discussion - Generated Dataset
I Simulation needed to overcome lack of data
I General problem when using deep networks on light field data
I Geometric objects are biologically not plausible
I Real data contains noise
I Realistic simulation with unrealistic scene
Chen, Krämer, Klein, Romen – Data Innovation Lab 45
Department of MathematicsTechnical University of Munich
Discussion - Generated Dataset
I Simulation needed to overcome lack of data
I General problem when using deep networks on light field data
I Geometric objects are biologically not plausible
I Real data contains noise
I Realistic simulation with unrealistic scene
Chen, Krämer, Klein, Romen – Data Innovation Lab 45
Department of MathematicsTechnical University of Munich
Discussion - Generated Dataset
I Simulation needed to overcome lack of data
I General problem when using deep networks on light field data
I Geometric objects are biologically not plausible
I Real data contains noise
I Realistic simulation with unrealistic scene
Chen, Krämer, Klein, Romen – Data Innovation Lab 45
Department of MathematicsTechnical University of Munich
Discussion - Generated Dataset
I Simulation needed to overcome lack of data
I General problem when using deep networks on light field data
I Geometric objects are biologically not plausible
I Real data contains noise
I Realistic simulation with unrealistic scene
Chen, Krämer, Klein, Romen – Data Innovation Lab 45
Department of MathematicsTechnical University of Munich
Discussion - Generated Dataset
I Simulation needed to overcome lack of data
I General problem when using deep networks on light field data
I Geometric objects are biologically not plausible
I Real data contains noise
I Realistic simulation with unrealistic scene
Chen, Krämer, Klein, Romen – Data Innovation Lab 45
Department of MathematicsTechnical University of Munich
Discussion - Light Field Data
I Typical light field resolutions in research papers:I Spatial resolution of 512× 512
I Angular resolution of 3× 3, 5× 5 or 9× 9
I e.g. see EPINET[9], VommaNet[8]
I Light field microscopy data in our project:
I Spatial resolution of 43× 43
I Angular resolution of 23× 23
Chen, Krämer, Klein, Romen – Data Innovation Lab 46
Department of MathematicsTechnical University of Munich
Discussion - Light Field Data
I Typical light field resolutions in research papers:I Spatial resolution of 512× 512
I Angular resolution of 3× 3, 5× 5 or 9× 9
I e.g. see EPINET[9], VommaNet[8]
I Light field microscopy data in our project:I Spatial resolution of 43× 43
I Angular resolution of 23× 23
Chen, Krämer, Klein, Romen – Data Innovation Lab 46
Department of MathematicsTechnical University of Munich
Discussion - Light Field Data
I Depth predictions with a resolution of 43× 43 too small
I Upsampling of spatial resolution to 256× 256
I Di�erences in baseline and depth range
I Architectures proposed in research papers were not directlysuited for our data
Chen, Krämer, Klein, Romen – Data Innovation Lab 47
Department of MathematicsTechnical University of Munich
Discussion - Light Field Data
I Depth predictions with a resolution of 43× 43 too small
I Upsampling of spatial resolution to 256× 256
I Di�erences in baseline and depth range
I Architectures proposed in research papers were not directlysuited for our data
Chen, Krämer, Klein, Romen – Data Innovation Lab 47
Department of MathematicsTechnical University of Munich
Discussion - Light Field Data
I Depth predictions with a resolution of 43× 43 too small
I Upsampling of spatial resolution to 256× 256
I Di�erences in baseline and depth range
I Architectures proposed in research papers were not directlysuited for our data
Chen, Krämer, Klein, Romen – Data Innovation Lab 47
Department of MathematicsTechnical University of Munich
Discussion - Light Field Data
I Depth predictions with a resolution of 43× 43 too small
I Upsampling of spatial resolution to 256× 256
I Di�erences in baseline and depth range
I Architectures proposed in research papers were not directlysuited for our data
Chen, Krämer, Klein, Romen – Data Innovation Lab 47
Department of MathematicsTechnical University of Munich
Outlook
I Realistic scene generation
I Networks that are invariant to camera configurations
I Light field reconstruction using depth prediction
Chen, Krämer, Klein, Romen – Data Innovation Lab 48
Department of MathematicsTechnical University of Munich
Outlook
I Realistic scene generation
I Networks that are invariant to camera configurations
I Light field reconstruction using depth prediction
Chen, Krämer, Klein, Romen – Data Innovation Lab 48
Department of MathematicsTechnical University of Munich
Outlook
I Realistic scene generation
I Networks that are invariant to camera configurations
I Light field reconstruction using depth prediction
Chen, Krämer, Klein, Romen – Data Innovation Lab 48
Department of MathematicsTechnical University of Munich
Summary
1 Introduction
2 Data
3 Networks
4 Results
5 Discussion
Chen, Krämer, Klein, Romen – Data Innovation Lab 49
Department of MathematicsTechnical University of Munich
Thank you for your a�ention
�estions?
Chen, Krämer, Klein, Romen – Data Innovation Lab 50
Department of MathematicsTechnical University of Munich
Bibliography I
[1] h�ps://www.rpcphotonics.com/product/mla-s100-f12/.
[2] Nearest neighbour interpolation.
[3] M. Broxton, L. Grosenick, S. Yang, N. Cohen, A. Andalman, K. Deisseroth, and M. Levoy.
Wave optics theory and 3-d deconvolution for the light field microscope.
Optics express, 21(21):25418–25439, 2013.
[4] X. Duan, X. Ye, Y. Li, and H. Li.
High quality depth estimation from monocular images based on depth prediction andenhancement sub-networks.
In 2018 IEEE International Conference on Multimedia and Expo (ICME), pages 1–6. IEEE,2018.
Chen, Krämer, Klein, Romen – Data Innovation Lab 51
Department of MathematicsTechnical University of Munich
Bibliography II
[5] L. D. Ellis, E. C. Soo, J. C. Achenbach, M. G. Morash, and K. H. Soanes.
Use of the zebrafish larvae as a model to study cigare�e smoke condensate toxicity.
PLoS One, 9(12):e115305, 2014.
[6] C. Hahne, A. Aggoun, V. Velisavljevic, S. Fiebig, and M. Pesch.
Baseline and triangulation geometry in a standard plenoptic camera.
International Journal of Computer Vision, 126(1):21–35, Jan 2018.
[7] I. Laina, C. Rupprecht, V. Belagiannis, F. Tombari, and N. Navab.
Deeper depth prediction with fully convolutional residual networks.
In 2016 Fourth international conference on 3D vision (3DV), pages 239–248. IEEE, 2016.
[8] H. Ma, H. Li, Z. Qian, S. Shi, and T. Mu.
Vommanet: an end-to-end network for disparity estimation from reflective and texture-less light field images.
arXiv preprint arXiv:1811.07124, 2018.
Chen, Krämer, Klein, Romen – Data Innovation Lab 52
Department of MathematicsTechnical University of Munich
Bibliography III
[9] C. Shin, H.-G. Jeon, Y. Yoon, I. So Kweon, and S. Joo Kim.
Epinet: A fully-convolutional neural network using epipolar geometry for depth fromlight field images.
In Proceedings of the IEEE Conference on Computer Vision and Pa�ern Recognition, pages4748–4757, 2018.
[10] C. Szegedy, S. Io�e, V. Vanhoucke, and A. A. Alemi.
Inception-v4, inception-resnet and the impact of residual connections on learning.
In Thirty-First AAAI Conference on Artificial Intelligence, 2017.
Chen, Krämer, Klein, Romen – Data Innovation Lab 53
Department of MathematicsTechnical University of Munich
Backup Slides
Chen, Krämer, Klein, Romen – Data Innovation Lab 54
Department of MathematicsTechnical University of Munich
Depth Estimate
Figure 22: Example of a depth estimate.[4]
Chen, Krämer, Klein, Romen – Data Innovation Lab 55
Department of MathematicsTechnical University of Munich
Bad Pixel Ratio
BPR(x, x̂; δ) =1
NM
N∑i=0
M∑j=0
I[δ < |xi,j − x̂ i,j|]
Chen, Krämer, Klein, Romen – Data Innovation Lab 56
Department of MathematicsTechnical University of Munich
Upsampling Strategies Published at IEEE International Conference on 3D Vision (3DV) 2016
Figure 2. From up-convolutions to up-projections. (a)Standard up-convolution. (b) The equivalent but fasterup-convolution. (c) Our novel up-projection block, fol-lowing residual logic. (d) The faster equivalent versionof (c)
We further extend simple up-convolutions usinga similar but inverse concept to [7] to create up-sampling res-blocks. The idea is to introduce asimple 3 × 3 convolution after the up-convolutionand to add a projection connection from the lowerresolution feature map to the result, as shown inFig. 2(c). Because of the different sizes, the small-sized map needs to be up-sampled using anotherup-convolution in the projection branch, but sincethe unpooling only needs to be applied once forboth branches, we just apply the 5 × 5 convolu-tions separately on the two branches. We call thisnew up-sampling block up-projection since it ex-tends the idea of the projection connection [7] toup-convolutions. Chaining up-projection blocks al-lows high-level information to be more efficientlypassed forward in the network while progressivelyincreasing feature map sizes. This enables the con-struction of our coherent, fully convolutional net-work for depth prediction. Fig. 2 shows the dif-ferences between an up-convolutional block to up-projection block. It also shows the correspondingfast versions that will be described in the followingsection.
Fast Up-Convolutions. One further contributionof this work is to reformulate the up-convolutionoperation so to make it more efficient, leading to adecrease of training time of the whole network of
Figure 3. Faster up-convolutions. Top row: the com-mon up-convolutional steps: unpooling doubles a fea-ture map’s size, filling the holes with zeros, and a 5 × 5convolution filters this map. Depending on the positionof the filter, only certain parts of it (A,B,C,D) are mul-tiplied with non-zero values. This motivates convolv-ing the original feature map with the 4 differently com-posed filters (bottom part) and interleaving them to ob-tain the same output, while avoiding zero multiplications.A,B,C,D only mark locations and the actual weight val-ues will differ
around 15%. This also applies to the newly intro-duced up-projection operation. The main intuitionis as follows: after unpooling 75% of the result-ing feature maps contain zeros, thus the following5 × 5 convolution mostly operates on zeros whichcan be avoided in our modified formulation. Thiscan be observed in Fig. 3. In the top left the origi-nal feature map is unpooled (top middle) and thenconvolved by a 5 × 5 filter. We observe that in anunpooled feature map, depending on the location(red, blue, purple, orange bounding boxes) of the5×5 filter, only certain weights are multiplied withpotentially non-zero values. These weights fall intofour non-overlapping groups, indicated by differentcolors and A,B,C,D in the figure. Based on the fil-ter groups, we arrange the original 5 × 5 filter tofour new filters of sizes (A) 3 × 3, (B) 3 × 2, (C)2× 3 and (D) 2× 2. Exactly the same output as theoriginal operation (unpooling and convolution) cannow be achieved by interleaving the elements of thefour resulting feature maps as in Fig. 3. The corre-sponding changes from a simple up-convolutionalblock to the proposed up-projection are shown inFig. 2 (d).
Figure 23: Di�erent up-sampling strategies[7]
Chen, Krämer, Klein, Romen – Data Innovation Lab 57
Department of MathematicsTechnical University of Munich
Upsampling StrategiesPublished at IEEE International Conference on 3D Vision (3DV) 2016
Figure 2. From up-convolutions to up-projections. (a)Standard up-convolution. (b) The equivalent but fasterup-convolution. (c) Our novel up-projection block, fol-lowing residual logic. (d) The faster equivalent versionof (c)
We further extend simple up-convolutions usinga similar but inverse concept to [7] to create up-sampling res-blocks. The idea is to introduce asimple 3 × 3 convolution after the up-convolutionand to add a projection connection from the lowerresolution feature map to the result, as shown inFig. 2(c). Because of the different sizes, the small-sized map needs to be up-sampled using anotherup-convolution in the projection branch, but sincethe unpooling only needs to be applied once forboth branches, we just apply the 5 × 5 convolu-tions separately on the two branches. We call thisnew up-sampling block up-projection since it ex-tends the idea of the projection connection [7] toup-convolutions. Chaining up-projection blocks al-lows high-level information to be more efficientlypassed forward in the network while progressivelyincreasing feature map sizes. This enables the con-struction of our coherent, fully convolutional net-work for depth prediction. Fig. 2 shows the dif-ferences between an up-convolutional block to up-projection block. It also shows the correspondingfast versions that will be described in the followingsection.
Fast Up-Convolutions. One further contributionof this work is to reformulate the up-convolutionoperation so to make it more efficient, leading to adecrease of training time of the whole network of
Figure 3. Faster up-convolutions. Top row: the com-mon up-convolutional steps: unpooling doubles a fea-ture map’s size, filling the holes with zeros, and a 5 × 5convolution filters this map. Depending on the positionof the filter, only certain parts of it (A,B,C,D) are mul-tiplied with non-zero values. This motivates convolv-ing the original feature map with the 4 differently com-posed filters (bottom part) and interleaving them to ob-tain the same output, while avoiding zero multiplications.A,B,C,D only mark locations and the actual weight val-ues will differ
around 15%. This also applies to the newly intro-duced up-projection operation. The main intuitionis as follows: after unpooling 75% of the result-ing feature maps contain zeros, thus the following5 × 5 convolution mostly operates on zeros whichcan be avoided in our modified formulation. Thiscan be observed in Fig. 3. In the top left the origi-nal feature map is unpooled (top middle) and thenconvolved by a 5 × 5 filter. We observe that in anunpooled feature map, depending on the location(red, blue, purple, orange bounding boxes) of the5×5 filter, only certain weights are multiplied withpotentially non-zero values. These weights fall intofour non-overlapping groups, indicated by differentcolors and A,B,C,D in the figure. Based on the fil-ter groups, we arrange the original 5 × 5 filter tofour new filters of sizes (A) 3 × 3, (B) 3 × 2, (C)2× 3 and (D) 2× 2. Exactly the same output as theoriginal operation (unpooling and convolution) cannow be achieved by interleaving the elements of thefour resulting feature maps as in Fig. 3. The corre-sponding changes from a simple up-convolutionalblock to the proposed up-projection are shown inFig. 2 (d).
Figure 24: Faster up-convolution[7]
Chen, Krämer, Klein, Romen – Data Innovation Lab 58
Department of MathematicsTechnical University of Munich
Upsampling Strategies
I Deconvolution can cause checkerboard artifacts
Figure 25: Nearest neighbor upsampling[2]
Chen, Krämer, Klein, Romen – Data Innovation Lab 59
Department of MathematicsTechnical University of Munich
Lenslet Network - Fish Eye - Fewer Epochs
Figure 26: Spheres prediction (bestepoch)
Figure 27: Spheres prediction (lastepoch)
Chen, Krämer, Klein, Romen – Data Innovation Lab 60
Department of MathematicsTechnical University of Munich
Lenslet Network - Spheres - Fewer Epochs
Figure 28: Spheres prediction (bestepoch)
Figure 29: Spheres prediction (lastepoch)
Chen, Krämer, Klein, Romen – Data Innovation Lab 61