物体计数--Learning To Count Objects in Images

Learning To Count Objects in Images

NIPS

http://www.robots.ox.ac.uk/~vgg/research/counting/

Matlab code: http://www.robots.ox.ac.uk/~vgg/research/counting/index_org.html

本文是第一个提出用密度图来计数的文献。虽然没有用深度学习，但是为后面的研究指明了方向。

Visual Geometry Group University of Oxford 还是很厉害的。

计数示意图：

2 The Framework

2.1 Learning to Count

这里首先介绍了怎么把人工标记的 dot 变为密度图，这里提出使用 normalized 2D Gaussian kernel

2.2 The MESA distance

这里定义了一个距离函数，用于度量估计的密度图和真值密度图之间的差异性，相当于后来的损失函数

measures the mismatch between the ground truth and the estimated densities (the loss) and has a significant impact on the performance of the entire learning framework

2.3 Optimization

误差函数的优化变为一个 a convex quadratic program

3 Experiments

cell counting

people counting