How does MATLAB calculate average precision with the evaluateDetectionResults function?

Does the evaluateDetectionResults function estimate average precision as the area under the precision-recall curve (AUC)? Furthermore, are the precision values interpolated before the estimation process?

Thank you in advance!

Answer 1

[Rajani Mishra]

For learning about how to use evaluateDetectionResults() please refer to its documentation page provided at the end. You can refer to references in its documentation page for learning about how it calculates average precision. The internal implementation of the function can’t be shared.

Documentation link: https://www.mathworks.com/help/vision/ref/evaluatedetectionprecision.html?s_tid=answers_rc2-1_p4_Topic

[Alexandros Kellaris]

Dear Mr. Rajani,

Thank you for the response. Unfortunately, the documentation, to the best of my knowledge, does not provide the details I am looking for. It is not necessary to obtain the internal function code, but at least the calculation method the function uses. Without this information, our work cannot be published since it is impossible to clarify the methodology in detail.

I would therefore be grateful to anyone who can give a bit more detail about this function, with respect to the aspects raised in the original question.

Kind regards,

Alexandros

[David Schedl]

Hi,

for scientific publications it would be great to have some citeable references.

I didn't find any references in the documentation (online and within Matlab's help). When looking into the source code, however, I found this:

% References
% ----------
%   [1] C. D. Manning, P. Raghavan, and H. Schutze. An Introduction to
%   Information Retrieval. Cambridge University Press, 2008.
%
%   [2] D. Hoiem, Y. Chodpathumwan, and Q. Dai. Diagnosing error in
%   object detectors. In Proc. ECCV, 2012.
%
%   [3] Dollar, Piotr, et al. "Pedestrian Detection: An Evaluation of the
%   State of the Art." Pattern Analysis and Machine Intelligence, IEEE
%   Transactions on 34.4 (2012): 743 - 761.

When I further looked into the internal functions I also found this comment:

% ... Follows
% PASCAL VOC 2011 average precision metric. ...

So the VOC 2011 challenge paper should be the scientific work that describes how the average precision is calculated:

Everingham, M., Van Gool, L., Williams, C.K.I., Winn, J., Zisserman, A.: The pascal visual object classes (voc) challenge. International Journal of Computer Vision 88(2), 303 – 338 (2010)

[Alexandros Kellaris]

Dear David,

Thank you very much for the update, things seem to be clearer now!

Kind regards,

Alex