PointRend uses a subdivision strategy to adaptively select a - - PowerPoint PPT Presentation

▶
pointrend uses a subdivision strategy to adaptively
SMART_READER_LITE
LIVE PREVIEW

PointRend uses a subdivision strategy to adaptively select a - - PowerPoint PPT Presentation

Oct 08, 2023 173 likes •412 views

PointRend: Image Segmentation as Rendering (Alexander Kirillov, Yuxin Wu, Kaiming He, and Ross Girshick) PointRend uses a subdivision strategy to adaptively select a non-uniform set of points at which to compute labels. ELF: Embedded Localization

slide-1
SLIDE 1

PointRend: Image Segmentation as Rendering (Alexander Kirillov, Yuxin Wu, Kaiming He, and Ross Girshick)

PointRend uses a subdivision strategy to adaptively select a non-uniform set of points at which to compute labels.

slide-2
SLIDE 2

ELF: Embedded Localization of Features in Pre-Trained CNN

An Guoyuan 20184637

Benbihi, Assia and Geist, Matthieu and Pradalier ICCV 2019

slide-3
SLIDE 3

Content

  • 1. Background
  • 2. Method (Part 1) - Saliency Maps
  • 3. Method (part 2) – Feature Map Selection
  • 4. Review and Result
  • 5. Discussion
slide-4
SLIDE 4

1 Background

slide-5
SLIDE 5

requirement1

slide-6
SLIDE 6

requirement2

slide-7
SLIDE 7
  • Repeatable key points detector  Harris
  • Reliable descriptor  LoG; SIFT
slide-8
SLIDE 8

Reference:Hands-on Machine Learning with Scikit-Learn, Keras, and TensorFlow, 2nd ed., O'Reilly, 2019 https ://towardsdatascience.com/a-comprehensive-guide-to-convolutional-neural-networks-the-eli5-way-3bd2b1164a53

A milestone: Convolutional Neural Network How to use CNN in image search?

slide-9
SLIDE 9

We often use CNN as a descriptor

slide-10
SLIDE 10

Q: how could CNN be helpful for the detector?

ELF: Embedded Localization of Features in Pre-Trained CNN

Feature map

slide-11
SLIDE 11

2 Method (part 1) – Saliency Maps

  • During backpropagation, the gradient is helpful for detecting keypoints.
slide-12
SLIDE 12

Saliency map

http://research.sualab.com/assets/images/interpretable- machine-learning-overview-2/saliency-map-with- gradient-concept.png

  • During backpropagation, the gradient is helpful for detecting keypoints.
slide-13
SLIDE 13

Saliency map from feature map.

Image Feature Map Saliency Map

  • The correlation between the feature space

and image space

  • For every node in feature map, calculate th

e gradient to all the pixels in the image. Apply the correlation to the features specifically and generate a visualization in image space .

slide-14
SLIDE 14

Q: how to find the best feature map?

? ?

slide-15
SLIDE 15

3 Method (part 2) – Feature Map Selection

  • High level representation
  • High resolution localization information
slide-16
SLIDE 16

High level representation

  • To represent an image, the higher, the better.

High level representation Low level representation

slide-17
SLIDE 17

High resolution localization information

  • To find a accurate location,

the lower, the better.

Low High Low High Middle

Low level saliency maps activate pixels more accurately.

slide-18
SLIDE 18

Summary for feature map selection

Solution: Visually observe the highest level which provides accurate localization.

  • High level representation  the higher, the better
  • High resolution localization information  the lower, the better
slide-19
SLIDE 19

4 Review and Result

slide-20
SLIDE 20

Review

This paper focus on these parts

slide-21
SLIDE 21

Result

slide-22
SLIDE 22

5 Discussion

slide-23
SLIDE 23

Discussion

  • New directions
  • Harris on feature map.
  • Selecting the best feature map: SIFT-LoG
  • Main Contribution:
  • Feature map based saliency map
  • Only use pre-trained CNN