62
Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley

Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Embed Size (px)

Citation preview

Page 1: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Describing People: A Poselet-Based Approach to

Attribute Classification

Lubomir Bourdev1,2 Subhransu Maji1

Jitendra Malik1

1EECS U.C. Berkeley 2Adobe Systems Inc.

Page 2: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Goal: Extract attributes from images of people

Page 3: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Who has long hair?

Page 4: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Who has short pants?

Page 5: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Male or female?

Page 6: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Prior work on poselets and on

attributes

Page 7: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Prior work on Poselets• Introduced by [Bourdev and Malik, ICCV09]• Detection with poselets [Bourdev et al, ECCV10]• Applications

• Segmentation [Brox et al, ECCV10] [Maire et al, ICCV 11]• Actions [Yang et al, CVPR10] [Maji et al, CVPR11] [Yao et

al, ICCV11]• Human parsing [Wang et al, CVPR11]• Semantic contours [Hariharan et al, ICCV11]• Subordinate level categorization [Farrell et al, ICCV11]

Page 8: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Prior work on Poselets• Introduced by [Bourdev and Malik, ICCV09]• Detection with poselets [Bourdev et al, ECCV10]• Applications

• Segmentation [Brox et al, ECCV10] [Maire et al, ICCV 11]• Actions [Yang et al, CVPR10] [Maji et al, CVPR11] [Yao et

al, ICCV11]• Human parsing [Wang et al, CVPR11]• Semantic contours [Hariharan et al, ICCV11]• Subordinate level categorization [Farrell et al, ICCV11]

Page 9: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Prior work on Poselets• Introduced by [Bourdev and Malik, ICCV09]• Detection with poselets [Bourdev et al, ECCV10]• Applications

• Segmentation [Brox et al, ECCV10] [Maire et al, ICCV 11]• Actions [Yang et al, CVPR10] [Maji et al, CVPR11] [Yao et

al, ICCV11]• Human parsing [Wang et al, CVPR11]• Semantic contours [Hariharan et al, ICCV11]• Subordinate level categorization [Farrell et al, ICCV11]

Page 10: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Prior work on Poselets• Introduced by [Bourdev and Malik, ICCV09]• Detection with poselets [Bourdev et al, ECCV10]• Applications

• Segmentation [Brox et al, ECCV10] [Maire et al, ICCV 11]• Actions [Yang et al, CVPR10] [Maji et al, CVPR11] [Yao et

al, ICCV11]• Human parsing [Wang et al, CVPR11]• Semantic contours [Hariharan et al, ICCV11]• Subordinate level categorization [Farrell et al, ICCV11]

Page 11: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Prior work on Poselets• Introduced by [Bourdev and Malik, ICCV09]• Detection with poselets [Bourdev et al, ECCV10]• Applications

• Segmentation [Brox et al, ECCV10] [Maire et al, ICCV 11]• Actions [Yang et al, CVPR10] [Maji et al, CVPR11] [Yao et

al, ICCV11]• Human parsing [Wang et al, CVPR11]• Semantic contours [Hariharan et al, ICCV11]• Subordinate level categorization [Farrell et al, ICCV11]

Page 12: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Prior work on AttributesAttributes as intermediate

parts

Discovering attributes from text

Discovering attributes from images

Attributes from motion capture

Joint learning of classes &

attributes

Image retrieval with attributes

Attributes and actions

Active learning with attributes

Attributes of people

Gender attribute[Cottrell and Medcalfe, NIPS90] [Golomb et al, NIPS90] [Moghaddam & Yang, PAMI02] [Ferrari & Zisserman, NIPS07] [Kumar et al, ECCV08] [Gallagher and Chen, CVPR08] [Cao et al, ACM08] [Lampert et al, CVPR09] [Farhadi et al, CVPR 09] [Wang et al, BMVC09] [Wang and Forsyth, ICCV09] [Kumar et al, ICCV09] [Farhadi et al, CVPR10] [Berg et al, ECCV10] [Wang and Mori, ECCV10] [Sigal et al, ECCV10] [Branson el al, ECCV10] [Hwang et al, CVPR11] [Parikh and Grauman, CVPR11] [Douze et al, CVPR11] [Kovashka et al, ICCV11] [Liu et al, CVPR11] [Qiu et al, ICCV11] [Yao et al, ICCV11] [Dhar et al, CVPR11] [Parikh and Grauman, ICCV11] [Siddiquie et al, CVPR11]

Page 13: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Prior work on AttributesAttributes as intermediate

parts

Discovering attributes from text

Discovering attributes from images

Attributes from motion capture

Joint learning of classes &

attributes

Image retrieval with attributes

Attributes and actions

Active learning with attributes

Attributes of people

Gender attribute[Cottrell and Medcalfe, NIPS90] [Golomb et al, NIPS90] [Moghaddam & Yang, PAMI02] [Ferrari & Zisserman, NIPS07] [Kumar et al, ECCV08] [Gallagher and Chen, CVPR08] [Cao et al, ACM08] [Lampert et al, CVPR09] [Farhadi et al, CVPR 09] [Wang et al, BMVC09] [Wang and Forsyth, ICCV09] [Kumar et al, ICCV09] [Farhadi et al, CVPR10] [Berg et al, ECCV10] [Wang and Mori, ECCV10] [Sigal et al, ECCV10] [Branson el al, ECCV10] [Hwang et al, CVPR11] [Parikh and Grauman, CVPR11] [Douze et al, CVPR11] [Kovashka et al, ICCV11] [Liu et al, CVPR11] [Qiu et al, ICCV11] [Yao et al, ICCV11] [Dhar et al, CVPR11] [Parikh and Grauman, ICCV11] [Siddiquie et al, CVPR11]

Page 14: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Prior work on AttributesAttributes as intermediate

parts

Discovering attributes from text

Discovering attributes from images

Attributes from motion capture

Joint learning of classes &

attributes

Image retrieval with attributes

Attributes and actions

Active learning with attributes

Attributes of people

Gender attribute[Cottrell and Medcalfe, NIPS90] [Golomb et al, NIPS90] [Moghaddam & Yang, PAMI02] [Ferrari & Zisserman, NIPS07] [Kumar et al, ECCV08] [Gallagher and Chen, CVPR08] [Cao et al, ACM08] [Lampert et al, CVPR09] [Farhadi et al, CVPR 09] [Wang et al, BMVC09] [Wang and Forsyth, ICCV09] [Kumar et al, ICCV09] [Farhadi et al, CVPR10] [Berg et al, ECCV10] [Wang and Mori, ECCV10] [Sigal et al, ECCV10] [Branson el al, ECCV10] [Hwang et al, CVPR11] [Parikh and Grauman, CVPR11] [Douze et al, CVPR11] [Kovashka et al, ICCV11] [Liu et al, CVPR11] [Qiu et al, ICCV11] [Yao et al, ICCV11] [Dhar et al, CVPR11] [Parikh and Grauman, ICCV11] [Siddiquie et al, CVPR11]

Page 15: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Prior work on AttributesAttributes as intermediate

parts

Discovering attributes from text

Discovering attributes from images

Attributes from motion capture

Joint learning of classes &

attributes

Image retrieval with attributes

Attributes and actions

Active learning with attributes

Attributes of people

Gender attribute[Cottrell and Medcalfe, NIPS90] [Golomb et al, NIPS90] [Moghaddam & Yang, PAMI02] [Ferrari & Zisserman, NIPS07] [Kumar et al, ECCV08] [Gallagher and Chen, CVPR08] [Cao et al, ACM08] [Lampert et al, CVPR09] [Farhadi et al, CVPR 09] [Wang et al, BMVC09] [Wang and Forsyth, ICCV09] [Kumar et al, ICCV09] [Farhadi et al, CVPR10] [Berg et al, ECCV10] [Wang and Mori, ECCV10] [Sigal et al, ECCV10] [Branson el al, ECCV10] [Hwang et al, CVPR11] [Parikh and Grauman, CVPR11] [Douze et al, CVPR11] [Kovashka et al, ICCV11] [Liu et al, CVPR11] [Qiu et al, ICCV11] [Yao et al, ICCV11] [Dhar et al, CVPR11] [Parikh and Grauman, ICCV11] [Siddiquie et al, CVPR11]

Page 16: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Prior work on AttributesAttributes as intermediate

parts

Discovering attributes from text

Discovering attributes from images

Attributes from motion capture

Joint learning of classes &

attributes

Image retrieval with attributes

Attributes and actions

Active learning with attributes

Attributes of people

Gender attribute[Cottrell and Medcalfe, NIPS90] [Golomb et al, NIPS90] [Moghaddam & Yang, PAMI02] [Ferrari & Zisserman, NIPS07] [Kumar et al, ECCV08] [Gallagher and Chen, CVPR08] [Cao et al, ACM08] [Lampert et al, CVPR09] [Farhadi et al, CVPR 09] [Wang et al, BMVC09] [Wang and Forsyth, ICCV09] [Kumar et al, ICCV09] [Farhadi et al, CVPR10] [Berg et al, ECCV10] [Wang and Mori, ECCV10] [Sigal et al, ECCV10] [Branson el al, ECCV10] [Hwang et al, CVPR11] [Parikh and Grauman, CVPR11] [Douze et al, CVPR11] [Kovashka et al, ICCV11] [Liu et al, CVPR11] [Qiu et al, ICCV11] [Yao et al, ICCV11] [Dhar et al, CVPR11] [Parikh and Grauman, ICCV11] [Siddiquie et al, CVPR11]

Page 17: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Prior work on AttributesAttributes as intermediate

parts

Discovering attributes from text

Discovering attributes from images

Attributes from motion capture

Joint learning of classes &

attributes

Image retrieval with attributes

Attributes and actions

Active learning with attributes

Attributes of people

Gender attribute[Cottrell and Medcalfe, NIPS90] [Golomb et al, NIPS90] [Moghaddam & Yang, PAMI02] [Ferrari & Zisserman, NIPS07] [Kumar et al, ECCV08] [Gallagher and Chen, CVPR08] [Cao et al, ACM08] [Lampert et al, CVPR09] [Farhadi et al, CVPR 09] [Wang et al, BMVC09] [Wang and Forsyth, ICCV09] [Kumar et al, ICCV09] [Farhadi et al, CVPR10] [Berg et al, ECCV10] [Wang and Mori, ECCV10] [Sigal et al, ECCV10] [Branson el al, ECCV10] [Hwang et al, CVPR11] [Parikh and Grauman, CVPR11] [Douze et al, CVPR11] [Kovashka et al, ICCV11] [Liu et al, CVPR11] [Qiu et al, ICCV11] [Yao et al, ICCV11] [Dhar et al, CVPR11] [Parikh and Grauman, ICCV11] [Siddiquie et al, CVPR11]

Page 18: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Prior work on AttributesAttributes as intermediate

parts

Discovering attributes from text

Discovering attributes from images

Attributes from motion capture

Joint learning of classes &

attributes

Image retrieval with attributes

Attributes and actions

Active learning with attributes

Attributes of people

Gender attribute[Cottrell and Medcalfe, NIPS90] [Golomb et al, NIPS90] [Moghaddam & Yang, PAMI02] [Ferrari & Zisserman, NIPS07] [Kumar et al, ECCV08] [Gallagher and Chen, CVPR08] [Cao et al, ACM08] [Lampert et al, CVPR09] [Farhadi et al, CVPR 09] [Wang et al, BMVC09] [Wang and Forsyth, ICCV09] [Kumar et al, ICCV09] [Farhadi et al, CVPR10] [Berg et al, ECCV10] [Wang and Mori, ECCV10] [Sigal et al, ECCV10] [Branson el al, ECCV10] [Hwang et al, CVPR11] [Parikh and Grauman, CVPR11] [Douze et al, CVPR11] [Kovashka et al, ICCV11] [Liu et al, CVPR11] [Qiu et al, ICCV11] [Yao et al, ICCV11] [Dhar et al, CVPR11] [Parikh and Grauman, ICCV11] [Siddiquie et al, CVPR11]

Page 19: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Prior work on AttributesAttributes as intermediate

parts

Discovering attributes from text

Discovering attributes from images

Attributes from motion capture

Joint learning of classes &

attributes

Image retrieval with attributes

Attributes and actions

Active learning with attributes

Attributes of people

Gender attribute[Cottrell and Medcalfe, NIPS90] [Golomb et al, NIPS90] [Moghaddam & Yang, PAMI02] [Ferrari & Zisserman, NIPS07] [Kumar et al, ECCV08] [Gallagher and Chen, CVPR08] [Cao et al, ACM08] [Lampert et al, CVPR09] [Farhadi et al, CVPR 09] [Wang et al, BMVC09] [Wang and Forsyth, ICCV09] [Kumar et al, ICCV09] [Farhadi et al, CVPR10] [Berg et al, ECCV10] [Wang and Mori, ECCV10] [Sigal et al, ECCV10] [Branson el al, ECCV10] [Hwang et al, CVPR11] [Parikh and Grauman, CVPR11] [Douze et al, CVPR11] [Kovashka et al, ICCV11] [Liu et al, CVPR11] [Qiu et al, ICCV11] [Yao et al, ICCV11] [Dhar et al, CVPR11] [Parikh and Grauman, ICCV11] [Siddiquie et al, CVPR11]

Page 20: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Prior work on AttributesAttributes as intermediate

parts

Discovering attributes from text

Discovering attributes from images

Attributes from motion capture

Joint learning of classes &

attributes

Image retrieval with attributes

Attributes and actions

Active learning with attributes

Attributes of people

Gender attribute[Cottrell and Medcalfe, NIPS90] [Golomb et al, NIPS90] [Moghaddam & Yang, PAMI02] [Ferrari & Zisserman, NIPS07] [Kumar et al, ECCV08] [Gallagher and Chen, CVPR08] [Cao et al, ACM08] [Lampert et al, CVPR09] [Farhadi et al, CVPR 09] [Wang et al, BMVC09] [Wang and Forsyth, ICCV09] [Kumar et al, ICCV09] [Farhadi et al, CVPR10] [Berg et al, ECCV10] [Wang and Mori, ECCV10] [Sigal et al, ECCV10] [Branson el al, ECCV10] [Hwang et al, CVPR11] [Parikh and Grauman, CVPR11] [Douze et al, CVPR11] [Kovashka et al, ICCV11] [Liu et al, CVPR11] [Qiu et al, ICCV11] [Yao et al, ICCV11] [Dhar et al, CVPR11] [Parikh and Grauman, ICCV11] [Siddiquie et al, CVPR11]

Page 21: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Prior work on AttributesAttributes as intermediate

parts

Discovering attributes from text

Discovering attributes from images

Attributes from motion capture

Joint learning of classes &

attributes

Image retrieval with attributes

Attributes and actions

Active learning with attributes

Attributes of people

Gender attribute[Cottrell and Medcalfe, NIPS90] [Golomb et al, NIPS90] [Moghaddam & Yang, PAMI02] [Ferrari & Zisserman, NIPS07] [Kumar et al, ECCV08] [Gallagher and Chen, CVPR08] [Cao et al, ACM08] [Lampert et al, CVPR09] [Farhadi et al, CVPR 09] [Wang et al, BMVC09] [Wang and Forsyth, ICCV09] [Kumar et al, ICCV09] [Farhadi et al, CVPR10] [Berg et al, ECCV10] [Wang and Mori, ECCV10] [Sigal et al, ECCV10] [Branson el al, ECCV10] [Hwang et al, CVPR11] [Parikh and Grauman, CVPR11] [Douze et al, CVPR11] [Kovashka et al, ICCV11] [Liu et al, CVPR11] [Qiu et al, ICCV11] [Yao et al, ICCV11] [Dhar et al, CVPR11] [Parikh and Grauman, ICCV11] [Siddiquie et al, CVPR11]

Page 22: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Prior work on AttributesAttributes as intermediate

parts

Discovering attributes from text

Discovering attributes from images

Attributes from motion capture

Joint learning of classes &

attributes

Image retrieval with attributes

Attributes and actions

Active learning with attributes

Attributes of people

Gender attribute[Cottrell and Medcalfe, NIPS90] [Golomb et al, NIPS90] [Moghaddam & Yang, PAMI02] [Ferrari & Zisserman, NIPS07] [Kumar et al, ECCV08] [Gallagher and Chen, CVPR08] [Cao et al, ACM08] [Lampert et al, CVPR09] [Farhadi et al, CVPR 09] [Wang et al, BMVC09] [Wang and Forsyth, ICCV09] [Kumar et al, ICCV09] [Farhadi et al, CVPR10] [Berg et al, ECCV10] [Wang and Mori, ECCV10] [Sigal et al, ECCV10] [Branson el al, ECCV10] [Hwang et al, CVPR11] [Parikh and Grauman, CVPR11] [Douze et al, CVPR11] [Kovashka et al, ICCV11] [Liu et al, CVPR11] [Qiu et al, ICCV11] [Yao et al, ICCV11] [Dhar et al, CVPR11] [Parikh and Grauman, ICCV11] [Siddiquie et al, CVPR11]

Page 23: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Poselets for Attribute Classification

Page 24: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Male or female?

Page 25: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Gender recognition is easier if we factor out the

pose

Page 26: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Poselets

[Bourdev & Malik ICCV09]

Page 27: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Poselets

Examples may differ visually but have common semantics

Page 28: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

How do we train a poselet?

Page 29: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Finding correspondences at training time

Given part of a human pose

How do we find a similar pose configuration in the training set?

Page 30: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

We use keypoints to annotate the joints, eyes, nose, etc. of people

Left Hip

Left Shoulder

Finding correspondences at training time

Page 31: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Residual Error

Finding correspondences at training time

Page 32: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Training poselet classifiers

Residual Error:

0.15

0.20

0.10

0.35

0.15

0.85

1. Given a seed patch2. Find the closest patch for every

other person3. Sort them by residual error4. Threshold them

Page 33: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Training poselet classifiers

1. Given a seed patch2. Find the closest patch for every

other person3. Sort them by residual error4. Threshold them5. Use them as positive training

examples to train a linear SVM with HOG features

Page 34: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Attribute Classification Algorithm at Test Time

Page 35: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Goal: Extract attributes of this person

Page 36: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Goal: Extract attributes of this person

Target person boundsBounds of other nearby people

Input:

Page 37: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Step 1: Detect poselet activations

[Bourdev et al, ECCV10]

Page 38: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Step 2: Cluster the activations

[Bourdev et al, ECCV10]

Page 39: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Step 3: Predict person bounds

[Bourdev et al, ECCV10]

Page 40: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Step 4: Identify the correct cluster

Max-flow in bipartite graph

Page 41: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

PoseletActivations

Start with its poselet activations

Page 42: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Features

FeaturesFeatures

PoseletActivations

• Pyramid HOG• LAB histogram• Skin features

• Hands-skin• Legs-skin

Poseletpatch

B .* CSkinmask

Armsmask

Page 43: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

PoseletActivations

Features

Poselet-levelAttributeClassifiers

Attribute Classification Overview

Page 44: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

PoseletActivations

Features

Poselet-levelAttributeClassifiers

Person-levelAttributeClassifiers

Attribute Classification Overview

Page 45: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

PoseletActivations

Features

Poselet-levelAttributeClassifiers

Person-levelAttributeClassifiers

Context-levelAttributeClassifiers

Attribute Classification Overview

Page 46: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Results

Page 47: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Our dataset• Source: VOC 2010 trainval for Person + H3D

• ~8000 annotations (4000 train + 4000 test)

• 9 binary attributes specified by 5 independent annotators via AMT

• Ground truth label: If 4 of the 5 agree

• Dataset will be made publicly available

Page 48: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Visual search on our test set

“Female”

“Wears hat”

Page 49: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

“Has long hair”

“Wears glasses”

Page 50: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

“Wears shorts”

“Has long sleeves”

Page 51: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

“Doesn’t have long sleeves”

Page 52: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Our baseline• Canny-modulated HOG with SPM kernel [Lazebnik et al

CVPR06]

• To help the baseline trained separate SPM for four viewpoints:

• For each attribute we pick the best SPM as our baseline

Full view Head zoom Upper body Legs

Page 53: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Precision/recall on our test setLabel

frequency- - - -

___SPM

___No context

___FullModel

Page 54: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

State-of-the-art Gender Recognition

• We outperform Cognitec (top-notch face recognizer)

• We outperform any gender recognizer based on frontal faces (are there others?)• 61% of our test have frontal faces.• Even with perfect classification of frontal faces,

max AP=80.5% vs. our AP of 82.4%

Page 55: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Men most confused as women

Confusions

Women most confused as menbaseball hat

long hair

hair hidden

Page 56: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Short pants most confused to be long pants

Non-T-shirt most confused to be T-shirt

annotationerrors

Are these pants short?

wrong person

occlusion

Page 57: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Best poselets per attribute

Gender:

Long Hair:

Wears glasses:

Page 58: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

“A woman with long hair, glasses and long pants”(??)

We can describe a picture of a person

Page 59: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Conclusion

Page 60: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

How poselets help in high-level vision

The image is a complex function of the viewpoint, pose,

appearance, etc.

Poselets decouple pose and camera view

from appearance

Page 61: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

Google “poselets” to get:

• The set of published poselet papers• H3D data set + Matlab tools• Java3D annotation tool + video tutorial• Matlab code to detect people using poselets• Our latest trained poselets

Page 62: Describing People: A Poselet-Based Approach to Attribute Classification Lubomir Bourdev 1,2 Subhransu Maji 1 Jitendra Malik 1 1 EECS U.C. Berkeley 2 Adobe

“A man with short hair, glasses, short sleeves and shorts”

“A man with short hair and long sleeves”“A person with short hair,

no hat and long sleeves”

“A woman with long hair, glasses, short sleeves andlong pants”

“A person with long pants”

“A computer vision professor who likes machine learning”

Failure modePoselets website

http://eecs.berkeley.edu/~lbourdev/poselets

• The set of published poselet papers• H3D data set + Matlab tools• Java3D annotation tool + video tutorial• Matlab code to detect people using poselets• Our latest trained poselets