Count-ception: Counting by Fully Convolutional Redundant Counting - - PowerPoint PPT Presentation

▶
count ception counting by fully convolutional redundant
SMART_READER_LITE
LIVE PREVIEW

Count-ception: Counting by Fully Convolutional Redundant Counting - - PowerPoint PPT Presentation

Jul 01, 2023 336 likes •864 views

Count-ception: Counting by Fully Convolutional Redundant Counting Joseph Paul Cohen - MILA, University of Montreal Genevive Boucher - IRIC, University of Montreal Craig A. Glastonbury - BDI, University of Oxford Henry Z. Lo -

slide-1
SLIDE 1

Count-ception: Counting by Fully Convolutional Redundant Counting

Joseph Paul Cohen - MILA, University of Montreal Geneviève Boucher - IRIC, University of Montreal Craig A. Glastonbury - BDI, University of Oxford Henry Z. Lo - University of Massachusetts Boston Yoshua Bengio - MILA, University of Montreal

slide-2
SLIDE 2

Count what?

Cells Penguins Cars People Sea lions

slide-3
SLIDE 3

Cell growth studies

Treat cells with different compounds and

  • bserve proliferation
slide-4
SLIDE 4

Cell growth studies

Bachstetter, MW151 Inhibited IL-1? Levels after Traumatic Brain Injury with No Effect on Microglia Physiological Responses, PLOS ONE, 2017

slide-5
SLIDE 5

Cell growth studies

At the Cell Counter: THP-1 Cells, Molecular Devices https://www.moleculardevices.com/cell-counter-thp-1-cells

slide-6
SLIDE 6

Complicated cell structure MBM Dataset Bone marrow, H&E stained. Healthy cells Obtained from TCGA 44 images, 126 ± 33 cells

slide-7
SLIDE 7

Complicated cell structure MBM Dataset Bone marrow, H&E stained. Healthy cells Obtained from TCGA 44 images, 126 ± 33 cells

slide-8
SLIDE 8
slide-9
SLIDE 9

Cell counting = State of practice

1. Create binary segmentation image 2. Watershed segmentation 3. Isolate and count

slide-10
SLIDE 10

1. Create binary segmentation image 2. Watershed segmentation 3. Isolate and count

Cell counting = State of practice

slide-11
SLIDE 11

1. Create binary segmentation image 2. Watershed segmentation 3. Isolate and count

Cell counting = State of practice

slide-12
SLIDE 12

1. Create binary segmentation image 2. Watershed segmentation 3. Isolate and count

Cell counting = State of practice

slide-13
SLIDE 13

1. Create binary segmentation image 2. Watershed segmentation 3. Isolate and count

Cell counting = State of practice

slide-14
SLIDE 14
slide-15
SLIDE 15
  • W. Xie, J. A. Noble, and A. Zisserman, “Microscopy cell counting and detection with fully convolutional regression networks,” 2016.
  • V. Lempitsky and A. Zisserman, “Learning To Count Objects in Images,” 2010.

Cell counting

slide-16
SLIDE 16

Redundant Counting

Gaussian Kernel

[Lempitsky and Zisserman 2010]

Square kernel size matches the receptive field! Square Kernel

[Cohen et al. 2017]

slide-17
SLIDE 17

Receptive field

Cell counting = State of research

Gaussian Kernel

[Lempitsky and Zisserman 2010]

slide-18
SLIDE 18

Receptive Field

Small example

slide-19
SLIDE 19

0.5

Cell counting = State of research

Gaussian Kernel

[Lempitsky and Zisserman 2010]

slide-20
SLIDE 20

0.2

Cell counting = State of research

Gaussian Kernel

[Lempitsky and Zisserman 2010]

slide-21
SLIDE 21

0.0

Cell counting = State of research

Gaussian Kernel

[Lempitsky and Zisserman 2010]

slide-22
SLIDE 22

1.0

Cell counting = State of research

Gaussian Kernel

[Lempitsky and Zisserman 2010]

Square Kernel

[Cohen et al. 2017]

slide-23
SLIDE 23

1.0

Cell counting = State of research

Square Kernel

[Cohen et al. 2017]

slide-24
SLIDE 24

1.0

Cell counting = State of research

Square Kernel

[Cohen et al. 2017]

slide-25
SLIDE 25

Why not increase the variance of the gaussian?

σ = 1 σ = 8 σ = 64 σ = 32 σ = 16

slide-26
SLIDE 26

Why not increase the variance of the gaussian?

σ = 1 σ = 8 σ = 64 σ = 32 σ = 16

0.1

slide-27
SLIDE 27

Why not increase the variance of the gaussian?

σ = 1

slide-28
SLIDE 28

Why not increase the variance of the gaussian?

σ = 1

slide-29
SLIDE 29

Why not increase the variance of the gaussian?

σ = 1

slide-30
SLIDE 30

Why not increase the variance of the gaussian?

σ = 1

slide-31
SLIDE 31

Why not increase the variance of the gaussian?

σ = 1

slide-32
SLIDE 32
slide-33
SLIDE 33

Count-ception Architecture

slide-34
SLIDE 34

Count-ception Architecture

slide-35
SLIDE 35

Fully Convolutional Training

L1 regression error Effective batch size 82,082 patches No pooling or strides Easy calculation of receptive field!

slide-36
SLIDE 36
slide-37
SLIDE 37

Does redundant counting help?

Increasing the stride reduces the number of regression targets

slide-38
SLIDE 38

N = Number of train and validation samples

slide-39
SLIDE 39
slide-40
SLIDE 40

N = Number of train and validation samples

slide-41
SLIDE 41

Count-ception applied to tissue cells

Craig Glastonbury - Big Data Institute - University of Oxford

Challenges: + Adjoining neighbors + Complex cell structure + Few non-cell regions

slide-42
SLIDE 42

N = Number of train and validation samples

slide-43
SLIDE 43
slide-44
SLIDE 44

Counting fungal spores

slide-45
SLIDE 45

Kaggle sea lion challenge (37th place) Implemented by Robin Dinse (Universität Koblenz-Landau)

Count sea lions

slide-46
SLIDE 46
slide-47
SLIDE 47

Do you need to count things?

Source Code:

Joseph Paul Cohen joseph@josephpcohen.com arXiv: https://arxiv.org/abs/1703.08710 Site: https://github.com/ieee8023/countception Lasagne + Theano https://github.com/ieee8023/countception Karas https://github.com/fizzoo/countception-recreation TensorFlow https://github.com/rdinse/sea-lion-counter PyTorch https://github.com/rwightman/pytorch-countception-sealion

slide-48
SLIDE 48
slide-49
SLIDE 49

ShortScience.org

Joseph Paul Cohen Henry Z Lo Swami Iyer Supported by:

slide-50
SLIDE 50

What is it and why?

A platform for post-publication discussion with over 800 public summaries in machine learning written by the community!

Browse summaries for a paper Browse by venue

slide-51
SLIDE 51

Goal - Accelerate Science

  • Programmatic organization of summaries and notes
  • Speed up the literature review process
  • Increase the number of active researchers
  • Decrease the barriers to understand and improve on concepts

ShortScience.org ArXiv Conferences