Open Saltb0xApps opened 6 years ago
Thank you! I've just uploaded the trained people detector, it is the file called "object_detector.svm" :-) The README instructions have been updated regarding this matter as well.
Thank you so much! I'm working on something very similar for my college graduation project and this is really helpful.
Would it be possible for you to create a public link to the dataset images & training xml file too? I am considering to further improve the accuracy of the landmarking file by continuing the training process with a larger dataset, and maybe splitting it into 3 categories like full, upper & lower body. I'd love to contribute improvements too if i'm able to achieve better results.
Sorry for the late response, I was out of town all week.
I'm so glad to see this project is of great help for your college thesis. And I also appreciate that you are so willing to contribute further!
This afternoon I will create a link with the files you have requested, no problem at all :-) The approach you are suggesting sounds very interesting, I would love to see the results you achieve later on!
If you have any in-depth questions on the training process or parameters setup you can read our paper: http://www.cys.cic.ipn.mx/ojs/index.php/CyS/article/view/2365/2109 https://www.researchgate.net/publication/301544464_Single-Camera_Automatic_Landmarking_for_People_Recognition_with_an_Ensemble_of_Regression_Trees
Please don't forget to cite us! And if you are still in doubt or something seems unclear, don't hesitate to ask and just let me know.
Best regards!
Done!
I've uploaded all the images used in each stage in a zip named "Images". Inside there are four numbered directories, indicating the order in which they were used, but also an "extra_images" folder containing images that were not included in the original experiments and you might find very useful.
The "mydataset8.xml" file has the final configuration for the people detector trained with Dlib's imglab tool, while "training_with_body_landmarks.xml" and "testing_with_body_landmarks.xml" contain the hand-marked body shape images to train the predictor and its corresponding validation set.
I wish you great success and all the best in the tasks that lie before you.
Hey karla3jo, Thank you so much! I'll keep you posted about any progress i make in the future. Would it be possible for you to share your email where i can contact you regarding any discussions/questions about computer vision & dlib?
I've spent a better of my final year working on dlib & face landmarking, and would love to discuss it in depth with someone to further improve accuracy of the models i'm currently able to train.
You can contact me at karla.andrea.trejo@upc.edu.
I'm looking forward to discuss these topics with you :-)
Hey there, Really impressive work! would it be possible for you to share the Body detection .svm file too?