Crowd Counting Models

List of Crowd Counting Models

The table below shows the crowd counting models available.

Model

Documentation

CSRNet

model.csrnet

Benchmarks

Model Accuracy

The table below shows the performance of CSRNet obtained from the originalGitHub repo, using Mean Absolute Error (MAE) as themetric. The reported metrics are close to the results from theCSRNet paper.

Model

Type

Dataset

MAE

CSRNet

dense

ShanghaiTech Part A

65.92

sparse

ShanghaiTech Part B

11.01

Dataset

TheShanghaiTech dataset was used. It contains 1,198 annotatedimages split into 2 parts: Part A contains 482 images with highly congested scenes, while Part B contains 716images with relatively sparse crowd scenes.