Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Bound Tightening Network for Robust Crowd Counting

Qiming Wu

TL;DR

A novel Bound Tightening Network for Robust Crowd Counting is proposed, which consists of three parts: base model, smooth regularization module and certify bound module that propagate the interval bound through the base model and utilize the layer weights to guide the network learning.

Abstract

Crowd Counting is a fundamental topic, aiming to estimate the number of individuals in the crowded images or videos fed from surveillance cameras. Recent works focus on improving counting accuracy, while ignoring the certified robustness of counting models. In this paper, we propose a novel Bound Tightening Network (BTN) for Robust Crowd Counting. It consists of three parts: base model, smooth regularization module and certify bound module. The core idea is to propagate the interval bound through the base model (certify bound module) and utilize the layer weights (smooth regularization module) to guide the network learning. Experiments on different benchmark datasets for counting demonstrate the effectiveness and efficiency of BTN.

Bound Tightening Network for Robust Crowd Counting

TL;DR

A novel Bound Tightening Network for Robust Crowd Counting is proposed, which consists of three parts: base model, smooth regularization module and certify bound module that propagate the interval bound through the base model and utilize the layer weights to guide the network learning.

Abstract

Crowd Counting is a fundamental topic, aiming to estimate the number of individuals in the crowded images or videos fed from surveillance cameras. Recent works focus on improving counting accuracy, while ignoring the certified robustness of counting models. In this paper, we propose a novel Bound Tightening Network (BTN) for Robust Crowd Counting. It consists of three parts: base model, smooth regularization module and certify bound module. The core idea is to propagate the interval bound through the base model (certify bound module) and utilize the layer weights (smooth regularization module) to guide the network learning. Experiments on different benchmark datasets for counting demonstrate the effectiveness and efficiency of BTN.
Paper Structure (8 sections, 1 theorem, 9 equations, 1 figure, 3 tables)

This paper contains 8 sections, 1 theorem, 9 equations, 1 figure, 3 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Certify Bound Module
  4. Smooth Regularization Module
  5. Robustness Optimization.
  6. Experiments
  7. Experiment Setup
  8. Conclusion and future work

Key Result

Lemma 1

($L_2$ Norm Cases). Consider the $L_2$ norm bounded adversarial example $\{\tilde{x}: ||\tilde{x}-x||_2 \le \epsilon_2 \}$ and the notations in Theorem.theorem:norm duality, for any neuron $j$ of $z_1$, we have:

Figures (1)

  • Figure 1: Overview of our proposed crowd counting model. It consists of three parts: Smooth Regularization Term, Crowd Counting Base Model and Certify Bound Module.

Theorems & Definitions (2)

  • Lemma 1
  • proof