Extended Multi-Column Convolutional Neural Network for Crowd Counting

Implementation of PCM 2018 paper
Extended Multi-Column Convolutional Neural Network for Crowd Counting
using tensorflow

installation

1. Install tensorflow
2. git clone
3. download VGG16 model file (vgg16.npy) from https://github.com/machrisaa/tensorflow-vgg

data setup

All the data setup process follows the pytorch version implementation:
svishwa/crowdcount-mcnn

train

tensorflow:
run python3 train.py A(or B)
model is saved to modelA/ or modelB/

test

run python3 test.py A(or B)

(uncomment code containing heatmap in network.py to generate heatmap)

heatmap

actual: 1110

predicted: 1181