forwchen / acf

ACF person detector

[NOTE]

This repo is a part of: https://bitbucket.org/fdesmedt/openframework/overview

I made some modifications to it for my own use. And I used it a long time ago, code for training the model is lost ...

[Usage]

Compile with the standard cmake and make procedure.

Detect person in image and crop them using:

./ped_acf path/to/image/dir output/path

The arguments must be directories, not individual files.

Results will be shown on the screen, cropped and saved in the output path, as well as written to the detections.xml file.

[Citations]

• Open Framework for combined pedestrian detection F. De Smedt and T. Goedemé, VISAPP 2015, Berlin

• On-board real-time tracking of pedestrians on a UAV F. De Smedt, D. Hulens and T. Goedemé, EVW 2015, Boston, USA

[Common Bugs]

You may need to replace cv::vector with std::vector if using OpenCV 3.

It may also be necessary to add "using namespace std;" into the gpu.hpp file inside OpenCV.

acf's Issues

编译错误float area[Dets.size()];

你好，编译的时候在nms.cpp文件中出现错误：float area[Dets.size()];表达式计算结果不是常数
和float o = w * h / area[j];无法从float[‘函数’]转为float，请问是什么原因呢？你有没有遇到呢？

How to train the model?

How to train the model? using the matlab code and how to convert Matlab-trained models into useable format (xml),? Thanks!!!