Helenology / Paper_MineSafe

A Brief Taste

This repository provides Python codes for the working paper ``Detecting Miners’ Unsafe Acts Automatically by Density Estimation and Deep Learning".

🛠 Installation

• Run the following command to install all dependencies in a conda environment named pyflann and TensorFlow Probability.

  sh install.sh
  

• Note that TensorFlow Probability is strict with some specific TensorFlow version. PPlease see https://github.com/tensorflow/probability/releases to choose the right version to install TensorFlow Probability or upgrate TensorFlow.

🔭 Codes Overview

The repository consists of two parts: simulation studies and real data.

Part I. Simulation

The simulation results are in simulation folder.

Part II. Real Data

Models (for Comparison)

Part I. Model Information

Models	Code Source	More Info
RPCA	Codes of the RPCA method are modified from the repository dganguli/robust-pca. $\ell_1$ norm bug: see Error in L1-norm implementation #11	An interesting finding: I found one typo from the original manuscript of RPCA, whose threshold used in line 3 of Algorithm 1 is $\mu$. This should (or may) be a typo, which should be substituted by $1/\mu$.
CSPCA	Codes are written by me because the authors provided no relevant codes (although I tried to search on the Internet as well as email the authors).	Discoveries & Bugs: 【1】 This work claimed to have used Fast computation of the L1-principal component of real-valued data's fast algorithm for solving the $\ell_1$-PCA problem in (9)-(11) on page 353. The problem is that the fast algorithm provided only one (max) PC, while the CSPCA needs $d(\geq 1)$ PCs. Thus, to avoid the above problem, I directly used the RPCA algorithm instead of using the fast algorithm to solve the aforementioned $\ell_1$-PCA problem. 【2】 No algorithm is proposed by this work. Not to mention how to implement the TV minimization in this work.
SSSR	Codes are written by me because the authors provided no relevant codes (although I tried to search on the Internet as well as email the authors).	【1】SLIC algorithm for superpixels; 【2】Features including: LBP (Local Binary Patterns);
PI	Original codes in C language are provided by the author.

Part II. Model Path

Models	Path	Desciption
RPCA	MineSafe-2024/models/robust-pca-master/r_pca.py	Python to implement the robust PCA (RPCA) algorithm.

Datasets

Dataset	Source	More Info
Airport	xxx