Hierarchical Sparse FRAME Models

Experiment 2: Objec, part, and key point localization

Code and data

Contents
1: Object, part, and key point localization
2: Imprecision-recall curves
3: Quantitative evaluation
4: Reference

1. Object, part, and key point localization

(ground truth for key points/ our models / AOT / DPM)

Bear Cat Cougar Cow Deer Lion Tiger Wolf

Bear

Cat

Cougar

Cow

Deer

Lion

Tiger

Wolf

2. Imprecision-recall curves

Cat


Lion


Tiger


Wolf


Deer


Cougar


Cow


Bear


3. Quantitative evaluation

Table 1. Comparison of AUCs for localization of object, parts and key points

Tasks
Object
Part
Key point
ours
AOT
LSVM
ours
AOT
LSVM
ours
AOT
LSVM
cat
0.954
0.949
0.700
0.955
0.950
0.718
0.954
0.949
0.700
lion
0.879
0.842
0.834
0.908
0.856
0.830
0.907
0.857
0.834
tiger
0.954
0.948
0.744
0.956
0.950
0.744
0.954
0.948
0.744
wolf
0.857
0.774
0.741
0.888
0.826
0.750
0.887
0.825
0.741
deer
0.738
0.675
0.559
0.736
0.673
0.570
0.738
0.676
0.565
cougar
0.960
0.936
0.831
0.961
0.939
0.825
0.960
0.938
0.831
cow
0.757
0.549
0.663
0.762
0.546
0.670
0.763
0.556
0.673
bear
0.769
0.607
0.744
0.776
0.605
0.745
0.773
0.611
0.751
Avg.
0.859
0.785
0.727
0.868
0.793
0.732
0.867
0.795
0.730

4. Reference

[1] Felzenszwalb, P. F., Girshick, R. B., McAllester, D., & Ramanan, D. (2010) Object detection with discriminatively trained part-based models. IEEE transactions on pattern analysis and machine intelligence.
[2] Si, Z., & Zhu. S. C. (2013) Learning and-or templates for object recognition and detection. IEEE transactions on pattern analysis and machine intelligence.