Crowed Counting with Deep Learning

Jan 1, 0001
Code
The Million Man March, Washington, D.C., October 1995 was the focus of a large crowd counting dispute. from Wikipedia

Crowd counting or crowd estimating is a technique used to count or estimate the number of people in a crowd (Wikipedia). Nowadays with help of Deep Learning, we can do this task with neural networks from an image. In this project I implemented model of paper entitled “Towards Perspective-Free Object Counting with Deep Learning” by Daniel O˜noro-Rubio et al.

Preprocess:

after resizing and rescaling the coordinates of each point in the image and applyin a gussain filter in each coordination, the data is ready for training a neural network. below is a sample of training data:

training sample
train data sample

Training

  • I use ADAM Optimizer for training network and power 2 of norm1 distatnce as loss function.
  • Due to the small amount of data, augmentation methods have also been used to increase the data. The process of changing the value of the Loss function is available below
    loss function during training
    loss of network during training

Results

Below is the result of traning proposed network with data augmentation

output with augmentation
output of network

Deep Learning past
Amir Mesbah
Amir Mesbah
Master student in Computer Engineering (with a major in AI)

I am a graduated Master’s student from University of Tehran.