CS6501: Deep Learning for Visual Recognition Recognizing People in - - PowerPoint PPT Presentation

▶

cs6501 deep learning for visual recognition

CS6501: Deep Learning for Visual Recognition Recognizing People in - - PowerPoint PPT Presentation

Mar 12, 2023 565 likes •854 views

CS6501: Deep Learning for Visual Recognition Recognizing People in Images Todays Class Face Detection Face Matching - and any type of matching Pose estimation Face Detection Face Detection: Viola-Jones Face Detector circa 2001

slide-1

SLIDE 1

CS6501: Deep Learning for Visual Recognition

Recognizing People in Images

slide-2

SLIDE 2

Face Detection
Face Matching - and any type of matching
Pose estimation

Today’s Class

slide-3

SLIDE 3

Face Detection

slide-4

SLIDE 4

Face Detection: Viola-Jones Face Detector circa 2001

1. Compute these types of features

across the image

2. Use a shallow classifier – e.g. ADA Boost
3. Non-Max Supression

slide-5

SLIDE 5

Face Detection: Any Object Detector

https://towardsdatascience.com/faced-cpu-real-time-face-detection-using-deep-learning-1488681c1602

slide-6

SLIDE 6

Face Detection can be Hard

WIDER FACE dataset.

slide-7

SLIDE 7

Person Identification: Simplest Case

Classify Among k-people in your database

slide-8

SLIDE 8

Face Matching and just Matching Things

Are these pairs of images, instances of the same?

slide-9

SLIDE 9

Matching Things: Siamese Networks

Chopra, Hadsell, and LeCun. Learning a Similarity Metric Discriminatively, with Application to FaceVerification

Find a neural network such that if two instances of the same thing are fed into the network, the outputs are similar under some simple distance metric.

Also called the embedding problem

slide-10

SLIDE 10

Matching Things: Siamese Networks

https://arxiv.org/pdf/1503.03832v1.pdf FaceNet: A Unified Embedding for Face Recognition and Clustering

!" !# $(!") $(!#)

slide-11

SLIDE 11

Matching Things: Siamese Networks

https://arxiv.org/pdf/1503.03832v1.pdf FaceNet: A Unified Embedding for Face Recognition and Clustering

!" !# $(!") $(!#) |$ !" − $ !# |

if x1 and x2 are the same person then minimize:

slide-12

SLIDE 12

Matching Things: Siamese Networks

https://arxiv.org/pdf/1503.03832v1.pdf FaceNet: A Unified Embedding for Face Recognition and Clustering

!" !# $(!") $(!#) |$ !" − $ !# |

if x1 and x2 are the same person then minimize: Beware of Trivial Solutions!

slide-13

SLIDE 13

Matching Things: Siamese Networks

https://arxiv.org/pdf/1503.03832v1.pdf FaceNet: A Unified Embedding for Face Recognition and Clustering

!" !# $(!") $(!#) −|$ !" − $ !# |

if x1 and x3 are not the same person then minimize:

slide-14

SLIDE 14

Better Idea: Triplet Loss. e.g. FaceNet

https://arxiv.org/pdf/1503.03832v1.pdf FaceNet: A Unified Embedding for Face Recognition and Clustering

!(#$) !(#&) !(#')

Minimize the following loss for every possible triplets

∑( ! #$ − ! #& − ! #$ − ! #' + +)

slide-15

SLIDE 15

Better Idea: Select Triplets that are Hard

https://arxiv.org/pdf/1503.03832v1.pdf FaceNet: A Unified Embedding for Face Recognition and Clustering

!(#$) !(#&) !(#')

∑( ! #$ − ! #& − ! #$ − ! #' + +)

Minimize the following loss for every possible triplets

slide-16

SLIDE 16

Pose Estimation

http://www.stat.ucla.edu/~xianjie.chen/projects/pose_estimation/pose_estimation.html

slide-17

SLIDE 17

Deep Pose

https://arxiv.org/pdf/1312.4659.pdf

slide-18

SLIDE 18

Deep Pose

https://arxiv.org/pdf/1312.4659.pdf

slide-19

SLIDE 19

Results

slide-20

SLIDE 20

Pose Model II: HourGlass Network

Hourglass Module

slide-21

SLIDE 21

Pose Model II: HourGlass Network

Hourglass Network

slide-22

SLIDE 22

Pose Model II: HourGlass Network

Hourglass Network

slide-23

SLIDE 23

Pose Model II: HourGlass Network

slide-24

SLIDE 24

Dense Pose

http://densepose.org/

slide-25

SLIDE 25

Dense Pose

http://densepose.org/

slide-26

SLIDE 26

Dense Pose

http://densepose.org/

slide-27

SLIDE 27

Dense Pose

http://densepose.org/

slide-28

SLIDE 28

Questions?

28