/mscnn

mscnn crowd counting model implementation, source from "Multi-scale Convolution Neural Networks for Crowd Counting" write by Zeng L, Xu X, Cai B, et al.

Primary LanguagePythonGNU General Public License v3.0GPL-3.0

HEAD

mscnn crowd counting model

======= License

Introduction

This is open source project for crowd counting. Implement with paper "Multi-scale Convolution Neural Networks for Crowd Counting" write by Zeng L, Xu X, Cai B, et al. For more details, please refer to arXiv paper

multi-scale block

mscnn_model

mscnn_architecture

result_display

result_table

Contents

  1. Installation
  2. Preparation
  3. Train/Eval
  4. Details

Installation

  1. Configuration requirements
python3.x

Please using GPU, suggestion more than GTX960

python-opencv
#tensorflow-gpu==1.0.0
#tensorflow==1.0.0
matplotlib==2.2.2
numpy==1.14.2

conda install -c https://conda.binstar.org/menpo opencv3
pip install -r requirements.txt
  1. Get the code
git clone https://github.com/Ling-Bao/mscnn
cd mscnn

Preparation

  1. ShanghaiTech Dataset. ShanghaiTech Dataset makes by Zhang Y, Zhou D, Chen S, et al. For more detail, please refer to paper "Single-Image Crowd Counting via Multi-Column Convolutional Neural Network" and click on here.

  2. Get dataset and its corresponding map label Baidu Yun Password: sags

  3. Unzip dataset to mscnn root directory

 tar -xzvf  Data_original.tar.gz

Train/Eval

Train is easy, just using following step.

  1. Train. Using mscnn_train.py to evalute mscnn model
python mscnn_train.py
  1. Eval. Using mscnn_eval.py to evalute mscnn model
python mscnn_eval.py

Details

  1. Improving model structure. Add Batch Normal after each convolution layer.

======= License

TAIL