Skip to content
Example train/test code for WIDER 2019 person search by languange track
Branch: master
Clone or download
Latest commit 078f569 May 15, 2019
Type Name Latest commit message Commit time
Failed to load latest commit information.
misc init commit May 15, 2019
.gitignore init commit May 15, 2019 init commit May 15, 2019
Retrieval.lua init commit May 15, 2019

Person Search by Language Description for WIDER Challenge 2019

This repo aims at providing an example train/test code for person search by language track in WIDER Face and Person Challenge 2019. The code is an implementation of Person Search with Natural Language Description, CVPR 2017. The code is modified slightly from the Person-Search-with-Natural-Language-Description to fit the new train/val split for the challenge and be able to output a result.txt file for evaluation.


This code is written in Lua and requires Torch. See the Torch installation documentation for more details. To run this code, the following packages must be installed:

Data Preparation

  1. Register the codalab account for WIDER challenge 2019 person search by language track and download the person search by lauguage train & val dataset.

  2. Save the downloaded trainval.json file to "data/trainval.json"

  3. Run as follows:

    python --input_json data/trainval.json --images_root data/imgs --max_length 50 --word_count_threshold 2 --output_json data/reidtalk.json --output_h5 data/reidtalk.h5

Testing on the validation set

  1. Download the pretrained model from snapshot and save it to "snapshot/lstm1_rnn512_bestACC.t7" .

  2. Run Retrieval.lua. The result should be around:

    top- 1 = 19.71%
    top- 5 = 43.20%
    top-10 = 55.24%


  1. Download the VGG-16 network pre-trained model from model and save all the download files to "model/".

  2. Run train.lua.


  title={Person search with natural language description},
  author={Li, Shuang and Xiao, Tong and Li, Hongsheng and Zhou, Bolei and Yue, Dayu and Wang, Xiaogang},
  journal={arXiv preprint arXiv:1702.05729},
You can’t perform that action at this time.