People Detection benchmark repository
Results
Author: Piotr Dollár.
Contact university/company: Microsoft, Washington, West Virginia, United States.
Contact email: pdollar@microsoft.com
Method name: Aggregate Channel Features (ACF): 2 variations according to the chosen trained person model: ACF Caltech and ACF Inria.
Description and pocessing time: The ACF detector proposes a very fast exhaustive search and a holistic person model using aggregate channel features. The ACF approach is implemented in Matlab and the computational cost is around 0.02 seconds per frame with 352x288 images. The tests have been performed on a Pentium IV with a CPU frequency of 2.4 GHz and 3GB RAM.
Parameters: Author default parameters here.
You can download the people detection results of this method by clicking here (Caltech) and here (Inria).
Results for each video sequence:
Video |
ACF |
ACF |
1 |
99.3 |
86.4 |
2 |
77.1 |
92.9 |
3 |
68.9 |
56.0 |
4 |
33.9 |
59.1 |
5 |
51.6 |
24.7 |
6 |
67.2 |
54.4 |
7 |
2.7 |
89.4 |
8 |
4.6 |
33.4 |
9 |
72.2 |
76.1 |
10 |
8.7 |
59.5 |
11 |
8.6 |
34.8 |
12 |
91.1 |
92.6 |
13 |
87.2 |
81.6 |
14 |
21.3 |
51.1 |
15 |
70.0 |
86.2 |
16 |
89.6 |
54.5 |
Average AUC |
53.4 |
64.5 |
Ranking |
3.38 |
2.25 |
Results for each background complexity:
Background complexity |
ACF |
ACF |
Baseline |
81.8 |
78.4 |
Dynamic Background |
42.8 |
41.9 |
Camera Jitter |
34.9 |
71.9 |
Intermittent Object Motion |
23.5 |
51.0 |
Shadow |
71.8 |
73.2 |
Average AUC |
51.0 |
63.3 |
Ranking |
2.80 |
2.00 |
Results for each classification complexity:
Classification complexity |
ACF |
ACF |
Low |
86.1 |
80.7 |
Medium |
64.4 |
55.3 |
High |
13.3 |
54.5 |
Average AUC |
54.6 |
63.5 |
Ranking |
2.33 |
2.33 |
