A deep learning model automatically counts the number of people in any image.
No internet connection or login is required—completely ad-free.
Counting people in a scene manually can be challenging.
Image Crowd Counter simplifies this by automating the counting process.
It functions entirely on-device, ensuring privacy and accessibility without the need for internet or login.
- Optimized for overhead views of crowds ranging from 10 to 500 people.
- This app may not function properly on older devices.
Contact: [email protected]
This is a personal project to build an iOS app that includes a deep learning model running on mobile devices.
Recent deep learning models have become increasingly large, making it challenging to run them on mobile devices.
I wondered, "Which tasks are useful when run on mobile devices and can be solved with moderately sized models?"
I discovered the task of crowd counting, which I might sometimes want to solve using my mobile devices, without any registration or internet connection.
The model included in this app is SASNet, introduced in the following paper: To Choose or to Fuse? Scale Selection for Crowd Counting (AAAI 2021)
Their implementation and trained model is publicly open in this repo under the Apache License Version 2.0 (license).
I converted their PyTorch model, which was already trained with the SanghaiTech dataset part B, into the Core ML package format using coremltools, and then wrote an iOS app with SwiftUI that provides the user interface for using the model.
The model accepts an input image with a size of 1024 x 768, so the app provides an image cropper that resizes images to that size.