search

FaceDetect / jointCascade_py

Joint Cascade Face Detection and Alignment
104 stars 59 forks source link

joint face detection #1

Closed denghuiru closed 9 years ago

denghuiru commented 9 years ago

Hi, I have researched face detection for a period of time and feel very happy that you can release the code of this paper. However, if I want to get a quick start of this code without training new model, I am not sure what to do, your README.txt also don't specify this. So wil you give a guidance about how to get a quick look of the result of this paper.

samuel1208 commented 9 years ago

Now it just support the training of that algorithm, and the test part is not supported now. Actually you can try to train the model with the readme step by step with AFW dataset(I advice you to train it on AFLW dataset which contain 20000+ faces). I'll release the test part code as soon as possible

denghuiru commented 9 years ago

Thanks very much for your answer, I will try to train the model step by step, and at same time looking forward to your test code. best regards!

samuel1208 commented 9 years ago

I have trained a model based on AFW(only contain 337 faces) with 5 non-face images, It just for demo. Please pull the latest code

denghuiru commented 9 years ago

Hi, I have run the demo_detect.py on the image pos.jpg, but, Why I can't see the detect result , it seems the code "img.show()" dose not execute. Besides, the run time for a image is so long about 177s, Why the paper said their run time is about 30ms for a image. Best regards!

samuel1208 commented 9 years ago

I have comment here

denghuiru commented 9 years ago

@samuel1208
Thanks to your analysis, We will survey the implementation process in detail. If you have made any progress in optimization, please let us konw. Thank you!

samuel1208 commented 9 years ago

@guo253 @denghuiru
I have optimized the global regression into trees. This get 10+ speedup, Now the detection speed is about 8s. I think it still have space to improve the speed. If you have any advises, please comment it here

denghuiru commented 9 years ago

@samuel1208
Hi, I have tested your new code on my computer, of course, the speed becomes so quickly, Next, I plan to research your code carefully. Besides, I believe if we change the code to C, it will improve the speed at least 10 times, maybe,we can try.

samuel1208 commented 9 years ago

Yes, but I want to optimize the train speed next. Especially the bootstrap part. I think achieve the a good benchmark on FDDB based on AFLW is more important now. Please let me know any of your progress. We can get more ideas via suggesting this project to more people.

denghuiru commented 9 years ago

@samuel1208 ok, I will deeply research this paper and once we each get some progress, we could let each other know!