Where is SVM classifier clue on this package?

[agn-7]

According to your published paper, the selected classifier is SVM and AdaBoost ([43] Kidono reference), but I cannot find SVM clue on your package.

I found the only AdaBoost model and its code (people_detector.cpp).

[koide3]

We tested several classifiers (SVM, AdaBoost, RandomForest) with Kidono's features. Since AdaBoost showed a bit better accuracy than the other ones, in this work, we are using AdaBoost instead of SVM.

[agn-7]

Do you have the SVM classifier version of this code?

[koide3]

Since I did it a few years ago, and I left my former university last year, I'm not sure if I can find the SVM model... If you want, I can share the dataset I used to train the classifiers so that you can train the model by yourself.

[agn-7]

Any help would be greatly appreciated. May you share it?

[agn-7]

Another question:

What are the extracted features on this project? And are these features the same in AdaBoost and SVM?

[koide3]

We are using the features proposed in the following paper: Kidono et al., Pedestrian Recognition Using High-definition LIDAR, IV, 2011

You can extract the features from a point cloud with this code. https://github.com/koide3/hdl_people_tracking/blob/master/include/hdl_people_detection/kidono_feature_extractor.hpp

[koide3]

Here, you can find the dataset for pedestrian detection. It consists of KITTI, sydney, and our own dataset. To avoid some copyright problems, I put only the extracted features saved in the libsvm format.

https://drive.google.com/open?id=1ovFA9gwjo1zD6cj_q0dnGFR6XYapxbxk

[agn-7]

Thanks a lot.

Now, can I use any classifier to train with this dataset?

[tkbtvrobert]

I would also like to know how to train on custom data(AdaBoost, RandomForest), any guidance to save hours reverse understanding the code would be great. Thank you very much!!

[koide3]

This package uses the standard cv::ml model, and you can replace it with any custom model you want to use. https://github.com/koide3/hdl_people_tracking/blob/master/src/kidono_human_classifier.cpp

[tkbtvrobert]

Very thank you for your advice, it helped me a lot.

[agn-7]

@koide3 Thanks for your response. However, I wish you had answered in 2019 :grinning:

[tkbtvrobert]

@koide3 Hello, I have another question to ask you, is the feature format of adaboost the same as libsvm? I want to train an adaboost model, how can I use kidono_feature_extractor.hpp to extract features? Thank you very much!!