Joint Face Detection and Alignment using Multi-task Cascaded Convolutional Neural Networks
Switch branches/tags
Nothing to show
Clone or download
CongWeilin Merge pull request #63 from sumsuddin/naming_and_type_fix
fixed minor inconsistencies on network layer names and type casting
Latest commit 2883feb Oct 1, 2018
Failed to load latest commit information.
12net Delete 12net.prototxt~ Mar 4, 2017
24net fix Mar 16, 2017
48net Delete 48net.prototxt~ Mar 4, 2017
demo fix bugs Mar 7, 2017
prepare_data path fix for directory generation Jul 12, 2018
.gitignore path fix for directory generation Jul 12, 2018 update Readme Mar 13, 2017
_config.yml Set theme jekyll-theme-cayman Jun 5, 2018


Joint Face Detection and Alignment using Multi-task Cascaded Convolutional Neural Networks.

This project provide you a method to update multi-task-loss for multi-input source.



  1. Ubuntu 14.04 or 16.04
  2. caffe && pycaffe:
  3. cPickle && cv2 && numpy

Train Data

The training data generate process can refer to Seanlinx/mtcnn

Sample almost similar to Seanlinx's can be found in prepare_data

  • step1. Download Wider Face Training part only from Official Website and unzip to replace WIDER_train

  • step2. Run to generate 12net training data. Besides, provide you an example to build imdb, Remember changing and adding new params.

  • step3. Run to generate hard sample. Run Combine these output and generate imdb.

  • step4. Similar to last step, Run to generate hard sample. Run Combine these output and generate imdb.

Strongly suggest readers generate training data themselves. The sample training data of 12net and 24net is available( Too big for Baidu Drive to upload 48net) by sending Email


The main idea is block backward propagation for different task

12net 12net 24net 24net 48net 48net


The Q&A bellow can solve most of your problem.

Q1: What data base do you use?
A1: Similar to official paper, Wider Face for detection and CelebA for alignment.

Q2: What is "12(24/48)net-only-cls.caffemodel" file for?
A2: Provide a initial weigh to train. Since caffe's initial weigh is random, a bad initial weigh may take a long ran to converge even might overfit before that.

Q3: Why preprocess images by minus 128?
A3: Separating data from (0,+) to (-,+), can make converge faster and more accurate. Refer to Batch Normalization Accelerating Deep Network Training by Reducing Internal Covariate Shift

Q4: Do you implement OHEM(Online-Hard-Example-Mining)?
A4: No. OHEM is used when training data is not that much big. Refer to faster-rcnn's writer RBG's paper

Q5: Ratio positive/negative samples for 12net?
A5: This caffemodel used neg:pos=3:1. Because 12net's function is to eliminate negative answers, similar to exclusive method, we should learn more about negative elininate the wrong answer.

Q6: Why your stride is different to official?
A6: If you input a (X,X) image, the output Y = (X-11)/2. Every point on output represent a ROI on input. The ROI's left side moving range = (0, X-12) on input, and (0, Y-1) on output. So that stride = (X-12)/(Y-1) ≈≈ 2 in this net.

Q7: What is roi(cls/pts).imdb used for?
A7: Use imdb can feed training data into training net faster. Instead of random search data from the hard-disk, reading data from a large file once to memory will save you a lot of time. imdb was created by python module-cPickle.

Q8: What is different from
A8: Matrix version use linear matrix to make calculation faster(160ms on FDDB). If you are green hand in this area, read Non-Matrix version to understand each process.

Q9: I want your training data for emergency use. How to use them? How to train? ...
A9: ???

Current Status

CongWeilin updated in 2017/3/5

Update to make calculate faster, about 160ms/image.