Crowd-Analysis-with-CNN

This programs is to estimate the limited number of people(~10) in small image and get density map which is explained below.
For its network, I modified WideResNet from here.

Guidance

For detailed info, please check each directory.
classification&regression - traim models which estimate a limited number of people (~10) in small size image by classification or regression
classification-test&regression-test - can test trained network with different dataset you used for training.
frame - explained later
cell - explained later
location - with output from num-of-ppl estimating network (i.e. network used in classification&regression), you can estimate exact coordinates of ppl in those imgs. This program is under developing.

Population estimation

classification&regression - estimate just number of people in cropped img by classification approach or regression approach

classification_test&regression_test - can test trained network with different dataset you used for training.

Density map

frame - create density map

cell - this is for comparison with frame.

video - create density map video.

Location estimation (under development)

location - with output from number of people estimating network (i.e. classification&regression), you can estimate exact coordinates of people in those imgs.

Requirements

・python3.7
・pytorch
conda install pytorch torchvision cudatoolkit=x.x -c pytorch