guoyilin / facedetection_cnn Goto Github PK

View Code? Open in Web Editor NEW

321.0 34.0 220.0 316 KB

Multi-view Face Detection Using Deep Convolutional Neural Networks

Python 98.58% Shell 1.42%

facedetection_cnn's Introduction

FaceDetection_CNN

Implement Yahoo Paper: Multi-view Face Detection Using Deep Convolutional Neural Networks

Image Preprocess aflw dataset[1]. Use iou>=0.5 as positive, iou<=0.3 as negative. You should download the aflw dataset by yourself.
Fine-tune Alex-Net using AFLW dataset. The model is in Baidu Yun: https://pan.baidu.com/s/1pJJ2WKN, or Google Drive: https://drive.google.com/file/d/0B8_dH3SiT7reMjJVRjJDXzJkRDQ/view?usp=sharing
Convert fully connected layers into convolutional layers by reshaping layer parameters, see [2], you can use the convert_full_conv() function in test.py for converting.
Get heat map for each scale of image.
Process heat map by using non-maximal suppression to accurately localize the faces.

========== Reference: [1]https://lrs.icg.tugraz.at/research/aflw/

[2] http://nbviewer.ipython.org/github/BVLC/caffe/blob/master/examples/net_surgery.ipynb

facedetection_cnn's People

Contributors

Stargazers

Watchers

Forkers

zjucsxxd anguoyang jwmneu chongbingbao litongyun cmxnono vzvzx facedetect zdltheone iwhm sunxingxingtf loae xieyanfu kli-casia vsooda uestcwangxiao hbliujun1111 bshawk richardhahahaha twnming zhouxiazx nianfudong chrisyang crycrane thuyang facegen pinguo-zhangyiwen mrgloom ggaoming zbxzc35 bikong2 xuguozhi likeucode lgyhero leezivin irvingshu runauto dan1900 lijian8 tdjgcy dlyshare lrtcwd joeking11829 flyinggh ilovecv silasxue lvchigo ndsl qsong4 paincomeback kimfeng ddrun arunpatala arasharchor siddharthaghosh seirouslee geshiming linzhineng johndpope clcarwin qaz734913414 iscas-lee takecareofbigboss panovr yuechengli apprisi wanjinchang xy280786378 zencoding liuhengli dejunzhang agarwalnaimish wang93 xiaohujecky ccsuzjj zeyuan1987 fireae justinbaby veterun facenx superyangwenwen apu1995 gezhiwei ronnie-tian ruudvannistelrooy karezi qijiezhao sunnyyouknow mydude ckrunauto honghucode hexiangquan chifang ctgushiwei caomw mvpduncan deepxkn kissyzhou junpeiz hxl1990

facedetection_cnn's Issues

No such file when running image_preprocess.py

I checked the path and made sure all the dependent files are loaded correctly. But it still shows "No such file"?

no test result

谢谢你分享的模型！
有个问题想问一下：
我的运行环境是windows（caffe，python）
最后保存的结果还是原图,没有标记的人脸box。
跟了一下程序，generateBoundingBox里的prob值非常小，true_boxes是空的。不知道为什么？
期待回复:)

what is lfw.txt in test.py and from where can i access that for testing the model??

After training the model, I get the loss to be very high(can you tell me how to decrease it).
Following is the output after 10000 iter of training that i got:

I0331 04:21:06.584506 2709 sgd_solver.cpp:273] Snapshotting solver state to binary proto file /media/jinesh/ACB6A202B6A1CCE0/Academics/8thSemester/CNN/train_iter_100000.solverstate
I0331 04:21:08.205349 2709 solver.cpp:311] Iteration 100000, loss = 0.00238461
I0331 04:21:08.205370 2709 solver.cpp:331] Iteration 100000, Testing net (#0)
I0331 04:22:55.422292 2709 solver.cpp:398] Test net output #0: accuracy = 0.993758
I0331 04:22:55.422312 2709 solver.cpp:316] Optimization Done.
I0331 04:22:55.464334 2709 caffe.cpp:259] Optimization Done.

Next for testing I opened test.py but it says:
if name == "main":
#convert_full_conv()
face_detection("lfw.txt")
so where is lfw.txt in the repository?

Pre-trained model

Do you have a pre-trained model on AFLW?

my test figure can not be drawed boxes. Can you help me, please?

There are no boxes in my test figure even I set the prob>=0.5. Maybe from the caffemodel, the probs are 0 or very little. But I don't know why? Can you help me, please?

Request for model from model in Baidu Yun

Hello!
I'm having trouble downloading model from Baidu Yun. Could it be possible to redirect me to another link or somehow send it to me?

Thank you in advance

Is the model available?

Can you share the Baidu SkyDrive link?

how can I run this algorithm

should I implement caffe first?

boxes not getting created while testing

I implemented the whole code but when we pass low resolution images or if number of faces are too many, this algorithm fails to detect any faces in the images for me .So is there a fault with my code or something else is a problem?Check this link for the sample image which is used for detecting when there are too many faces.https://freemedicalcliniclifesaver.files.wordpress.com/2012/01/2012-class-photo1.jpg
Thank you for giving me time on this.

How can I generate "aflw/crop_images/train.txt" and "aflw/crop_images/val.txt"

Hi, I finished image preprocessing step using provided code "image_preprocess.py" and now I want to train the model using Caffe. However, there is train.txt and val.txt for training session and I cannot figure out how to generate train.txt and val.txt. It seems "aflw.list" file does a related job but I'm not sure. Could anyone teach me how to generate train.txt and val.txt? Thanks

Hi, Can you explain please in which environments you tested your code, versions of librairies and all things could help us to implement easely.Thanks in advance.