blockchainlover / person_search

End-to-End Deep Learning for Person Search

Person Search Project

This repository hosts the code for our paper End-to-End Deep Learning for Person Search. The code is modified from the py-faster-rcnn written by Ross Girshick.

Requirements

1. Requirements for Caffe and pycaffe (see: Caffe installation instructions)

Note: Caffe must be built with support for Python layers!

# In your Makefile.config, make sure to have this line uncommented
WITH_PYTHON_LAYER := 1
# Unrelatedly, it's also recommended that you use CUDNN
USE_CUDNN := 1

You can download my Makefile.config for reference. 2. Python packages you might not have: Cython, python-opencv, easydict (>=1.6), PyYAML 3. MATLAB is required for processing the raw data.

Installation

1. Build the Cython modules

   cd lib
   make

2. Build Caffe and pycaffe

   cd caffe-fast-rcnn
   # Now follow the Caffe installation instructions here:
   #   http://caffe.berkeleyvision.org/installation.html
   
   # If you're experienced with Caffe and have all of the requirements installed
   # and your Makefile.config in place, then simply do:
   make -j8 && make pycaffe

Experiments

1. Request the dataset from sli [at] ee.cuhk.edu.hk (academic only)

2. Convert the raw dataset into formatted database

   scripts/make_db.sh /path/to/the/downloaded/dataset.zip

3. Pretraining

   data/scripts/fetch_imagenet_models.sh
   experiments/scripts/pretrain.sh

   Or you may directly download a pretrained model from here to output/psdb_pretrain/.

4. Training and evaluation

   experiments/scripts/train.sh 0

   Or you may directly download a trained model from here to output/psdb_train/, and test the performance with

   python2 tools/test_net.py --gpu 0 \
     --gallery_def models/psdb/VGG16/test_gallery.prototxt \
     --probe_def models/psdb/VGG16/test_probe.prototxt \
     --net output/psdb_train/VGG16_iter_100000.caffemodel \
     --cfg experiments/cfgs/train.yml \
     --imdb psdb_test

   The result should be around

   mAP: 55.79%
   top-1: 62.17%

Citation

@article{xiaoli2016end,
  title={End-to-End Deep Learning for Person Search},
  author={Xiao, Tong and Li, Shuang and Wang, Bochao and Lin, Liang and Wang, Xiaogang},
  journal={arXiv:1604.01850},
  year={2016}
}