lincolnhard / mtcnn-head-detection

Head detection with modified cascaded CNN

MTCNN head detection

Set up

Set up environment and copy C++ layer code to Caffe's source code tree.

$ export PYTHONPATH=/path/to/mtcnn-head-detection:$PYTHONPATH
$ export CAFFE_HOME=/path/to/caffe
$ pip install easydict
$ pip install lmdb
$ sh layers/copy.sh

Compile Caffe following its document.

Prepare data

Download dataset SCUT-HEAD. Unzip and put them in data directory.

Train

pnet

python jfda/prepare.py --net p --wider --worker 8
python jfda/train.py --net p --gpu 0 --size 128 --lr 0.05 --lrw 0.1 --lrp 5 --wd 0.0001 --epoch 25

rnet

Choose appropriate pnet caffemodel to generate prior for rnet, and edit cfg.PROPOSAL_NETS in config.py

python jfda/prepare.py --net r --gpu 0 --detect --wider --worker 4
python jfda/train.py --net r --gpu 0 --size 128 --lr 0.05 --lrw 0.1 --lrp 5 --wd 0.0001 --epoch 25

onet

Choose appropriate rnet caffemodel to generate prior for onet, and edit cfg.PROPOSAL_NETS in config.py

python jfda/prepare.py --net o --gpu 0 --detect --wider --worker 4
python jfda/train.py --net o --gpu $GPU --size 64 --lr 0.05 --lrw 0.1 --lrp 7 --wd 0.0001 --epoch 35

Test

python simpledemo.py

Note

1. Landmark alignment in original mtcnn is removed in this repo. Here only do object classification and bounding box regression.

2. Each convolutional layer kernel number in onet has reduced for faster network inference.

Result

pnet

rnet

onet

References

• A Convolutional Neural Network Cascade for Face Detection
• Joint Face Detection and Alignment using Multi-task Cascaded Convolutional Networks
• MTCNN_face_detection_alignment
• py-faster-rcnn
• opencv-mtcnn

mtcnn-head-detection's Issues

Can you provide the pre-trained model?

Unable to find cheat and brainwash dataset

haneesh@haneesh:~/Desktop/mtcnn-head-detection-master$ python3 jfda/prepare.py --net p --wider --worker 8
[2020-01-05 20:31:13,719][INFO] loading WIDER
parsing /home/haneesh/Desktop/mtcnn-head-detection-master/data/SCUT_HEAD_Part_B/JPEGImages/ ...
parsing /home/haneesh/Desktop/mtcnn-head-detection-master/data/SCUT_HEAD_Part_A/JPEGImages/ ...
parsing /media/lincolnhard/11e5b063-4031-49ed-bf24-312fc250e90c/lincoln/cheat/JPEGImages/ ...
Traceback (most recent call last):
File "jfda/prepare.py", line 544, in
gen_wider()
File "jfda/prepare.py", line 270, in gen_wider
train_data, val_data = load_scutbrainwashcheat()
File "/home/haneesh/Desktop/mtcnn-head-detection-master/jfda/utils.py", line 15, in load_scutbrainwashcheat
for name in os.listdir(trainimdir):
FileNotFoundError: [Errno 2] No such file or directory: '/media/lincolnhard/11e5b063-4031-49ed-bf24-312fc250e90c/lincoln/cheat/JPEGImages/'

data_layer.py 出错

您好，data_layer.py这个文件是不是错的？

IndexError: Index out of range

Hi, I have generated my own data,there has some issues like below when I trained,
I0906 03:52:38.055449 145076 layer_factory.hpp:77] Creating layer data
I0906 03:52:38.056452 145076 net.cpp:86] Creating Layer data
I0906 03:52:38.056471 145076 net.cpp:382] data -> data
I0906 03:52:38.056488 145076 net.cpp:382] data -> bbox_target
I0906 03:52:38.056497 145076 net.cpp:382] data -> label
Traceback (most recent call last):
File "jfda/train.py", line 123, in
solver = Solver(solver_prototxt, args)
File "jfda/train.py", line 60, in init
self.solver = caffe.SGDSolver(tmp_solver_prototxt)
File "/home/ll/Project/Object_Detection/mtcnn-head-detection/layers/data_layer.py", line 55, in setup
self.reshape(bottom, top)
File "/home/ll/Project/Object_Detection/mtcnn-head-detection/layers/data_layer.py", line 61, in reshape
top[3].reshape(self.n)
IndexError: Index out of range

I would like to cite your work

Hello!
Your implementation has greatly aided in my research!
I changed some things up in the code ( and will happily provide all changes upon request) and the model is currently being re-trained for purposes not related to facial recognition.
I would like to firstly thank you and show my appreciation.
I'm not sure how to cite your work other than putting a link to your github in my paper.
Other than that is there any special way you would like to be cited?
Thanks again!

在训练时遇到caffe.SGDSolver这个错误

I0704 15:17:48.250639 31549 net.cpp:382] data -> bbox_target
I0704 15:17:48.250646 31549 net.cpp:382] data -> label
Traceback (most recent call last):
File "jfda/train.py", line 125, in
solver = Solver(solver_prototxt, args)
File "jfda/train.py", line 62, in init
self.solver = caffe.SGDSolver(tmp_solver_prototxt)
File "/home/isp/workspace/opencvLab/mtcnn-head-detection/layers/data_layer.py", line 55, in setup
self.reshape(bottom, top)
File "/home/isp/workspace/opencvLab/mtcnn-head-detection/layers/data_layer.py", line 61, in reshape
top[3].reshape(self.n)
IndexError: Index out of range

inference time

想问下up主，caffe下跑单一图片的时间有测算过么?
谢谢~