北大选课网2021年春季验证码识别

Overview

北大选课网验证码识别 2021 年春季学期

Powered by Elector Quartet (@Rabbit, @xmcp, @SpiritedAwayCN, @gzz)

数据集描述

最初的数据集为 5130 张人工标记的验证码,之后利用早期训练好的模型在选课网上进行自动验证 (自举),又收集到 102741 张验证码,并对其中 2363 张识别错误的验证码进行人工标注。以上三部分合并为 GIF 验证码数据集

对 GIF 验证码数据集进行预处理,得到实际用于模型训练的数据集,它由裁剪好的单字母灰度图组成,图片尺寸 52x52,8 位色深,文件格式 PNG

训练集/测试集按照 8:2 进行划分,对训练集的数据进行数据增强,方式为旋转特定的几个角度,测试集不做特殊处理

在 const.py 中,DATA_RAW_DIR 对应最初的 5130 张人工标记的验证码,DATA_BOOTSTRAP_DIR 对应所有通过自举得到的验证码 (包含验证正确/错误的验证码),DATA_MANUALLY_LABELED_DIR 对应所有人工标注的验证码

数据集下载

百度网盘下载链接:https://pan.baidu.com/s/12GcfXYaLbHFC1WkzWhre2w (提取码:xsl3)

  1. 最初的 5130 张人工标注的验证码 xmcp_5130.zip,由 xmcp 提供,详见 xmcp/elective-dataset-2021spring
  2. 自举得到的验证码以 10000 张为单位,分为 11 个 bootstrap_all.x.zip
  3. 人工标注的验证码为 manually_labeled.zip,其中 manually_labeled_20210309/ 对应第一批加入到 CNN 模型训练的人工标注验证码,manually_labeled_20210310/manually_labeled_20210311/ 对应第二批加入到 CNN 模型训练的人工标注验证码
  4. 直接用于模型 210311_1 训练/测试的单字母灰度图 crop.20210311.1.tar.gz,共包含 354008 张单字母灰度图,对应于毫秒时间戳不超过 1615449512408 的 88502 张 GIF 验证码

图像预处理

基本采用 SpiritedAwayCN/ElectiveCaptCha 的方法进行预处理,相关算法的介绍参看 SpiritedAwayCN/ElectiveCaptCha 的说明

本项目对 SpiritedAwayCN/ElectiveCaptCha 的算法做了以下两点改进:

  1. 验证码的字母可能会变色,此时隔 4 帧作差,会有上一个字母的残留,因此加入 M_mask 的相关代码,每提取出一个字母,就将其加入到 M_mask 中,下一个字母提取的时候,会通过借助 M_mask 将之前已经提取出的字母对应的像素从当前字母的提取结果中抹去,以消除字母变色对后 3 个字母的提取所产生的影响
  2. 验证码的第一个字母可能在前 4 帧里颜色较浅,取第 4 帧转为灰度图后,第一个字母对应像素的灰度值大于全局平均灰度值,利用 OSTU 二值化时会被抹去,填充成白色,最后得到一张只剩下少量噪音残留的空白图像。改进的方法是,首先将 4 个关键帧的灰度图叠加,保留颜色最深的灰度值 (最小值),利用第 5-16 帧中第一个字母的颜色逐渐加深的特点,获得灰度值较深的包含第一个字母轮廓的 M_merge,此时再利用 OSTU 二值化,就基本上能留住第一个字母的信息,然后将 M_merge 的右三个字母抹去,仅保留包含第一个字母的轮廓,再根据此时的 M_merge,将第 1 个关键帧的灰度图中第一个字母对应像素的灰度值降低,此时再调用之前的处理函数,就可以更加准确地提取出第一个字母

模型训练与测试

以测试集的测试结果作为模型训练结果的参考,以模型在选课网上在线验证的结果作为模型的实际准确率

模型 测试集准确率 (单字母) 在线测试准确率 (四字母) / % 备注
210308_3 0.9968 (15818 / 15869) 0.9675 (17253 / 17832) 扩大数据集,进行数据增强
210309_2 0.9952 (33811 / 33973) 0.9799 (11659 / 11898) 扩大数据集,加入第一批人工标注的验证码
210309_3 0.9951 (33807 / 33973) 0.9788 (11847 / 12104) 同 210309_2,但数据增强的旋转角度范围缩小至 [-15, 15]
210309_4 0.9928 (33730 / 33973) 0.9727 (9527 / 9794) 同 210309_2,但不进行数据增强
210311_1 0.9977 (70467 / 70630) 0.9916 (10219 / 10306) 扩大数据集,加入第二批人工标注的验证码,优化第一个字母的图像处理算法
210311_2 0.9961 (70643 / 70918) 0.9838 (11230 / 11415) 同 210311_1,但使用未经优化的图像处理算法

目前准确率最高的模型为 210311_1,该模型的训练结果、训练时的终端输出、训练/测试的相关代码已在 dist/ 中提供

环境依赖

开发环境 Python 版本为 3.6.8,相关依赖包版本参考 requirements.txt,其中 PyTorch 版本应大于 1.4.x,否则无法读取 dist/ 中提供的模型

模型训练在阿里云服务器上进行,CUDA 版本 11.0,使用的 GPU 为 Tesla V100-SXM2-32GB,使用的 CPU 为 Intel(R) Xeon(R) Platinum 8163 CPU @ 2.50GHz,训练集一次性读入内存,以 210311_1 的训练/测试为例,数据量为 88502 张 GIF 验证码,数据增强的旋转角度范围为 [-30, 30],数据集扩增至 9 倍规模,占用的内存量为 32.9 GB,每个 Epoch 的耗时约为 309 秒

文件描述

preprocess.py

图像预处理的相关算法,实际有用的函数是 extract_c0_v1, extract_c0_v2, extract_c123, crop,处理流程参考 main 函数

cnn.py

与 CNN 模型相关的代码,包括数据集定义、数据增强、CNN 模型定义、训练集/测试集划分、模型训练、模型测试等

bootstrap.py

与模型自举相关的代码,在选课网上在线测试现有模型的准确率,并且将预测正确/错误的模型分类收集,用于扩增数据集

label_server.py

用于人工标注验证码的网站后端,前端代码在 web/ 中。提供图像预处理、输入检查等功能,为人工标注验证码提供方便,所使用的数据源为模型自举过程中收集到的识别错误的验证码

config.ini

与选课网账户相关的信息,包括学号、密码、是否为双学位账号,主要用于 bootstrap.py

致谢

Owner
Rabbit
CCME, Peking University
Rabbit
A 3-line lisp implementation

Nanolisp The download page of many a language harbors deep senses of forboding, of evil lurking in its native lair. You feel that the language is not

5 Jun 17, 2022
Turn crypto miner on/off depending on powerwall charge level

Mining Crypto with Tesla Solar and Powerwalls This script turns a crypto miner on and off when the Tesla Powerwall level drops/rises above a certain t

Matt 1 Nov 09, 2021
Syntax highlighting for yarn.lock and bun.lockb files

Yarn.lock Syntax Highlighting Syntax highlighting for yarn.lock and bun.lockb files Installation Plugin is not publushed yet on Package Control, to in

Alexander Kuznetsov 4 Jul 06, 2022
Gives criticality score for an open source project

Open Source Project Criticality Score (Beta) This project is maintained by members of the Securing Critical Projects WG. Goals Generate a criticality

Open Source Security Foundation (OpenSSF) 1.1k Dec 23, 2022
Easy, clean, reliable Python 2/3 compatibility

Overview: Easy, clean, reliable Python 2/3 compatibility python-future is the missing compatibility layer between Python 2 and Python 3. It allows you

Python Charmers 1.2k Jan 08, 2023
The LiberaPay archive module for the SeanPM life archive project.

By: Top README.md Read this article in a different language Sorted by: A-Z Sorting options unavailable ( af Afrikaans Afrikaans | sq Shqiptare Albania

Sean P. Myrick V19.1.7.2 1 Aug 26, 2022
Python script that automates the tasks involved in starting a new coding project

Auto Project Builder Automates the repetitive tasks while starting a new project Installation Use the REQUIREMENTS.txt file to install the dependencie

Prathap S S 1 Feb 03, 2022
Python-geoarrow - Storing geometry data in Apache Arrow format

geoarrow Storing geometry data in Apache Arrow format Installation $ pip install

Joris Van den Bossche 11 Mar 03, 2022
App to decide weekly winners in H2H 1 Win (9 Cat)

Fantasy Weekly Winner for H2H 1 Win (9 Cat) Yahoo Fantasy API Read

Sai Atmakuri 1 Dec 31, 2021
Project5 Data processing system

Project5-Data-processing-system User just needed to copy both these file to a folder and open Project5.py using cmd or using any python ide. It is to

1 Nov 23, 2021
Covid-ml-predictors - COVID predictions using AI.

COVID Predictions This repo contains ML models to be trained on COVID-19 data from the UK, sourced off of Kaggle here. This uses many different ML mod

1 Jan 09, 2022
RDFLib is a Python library for working with RDF, a simple yet powerful language for representing information.

RDFLib RDFLib is a pure Python package for working with RDF. RDFLib contains most things you need to work with RDF, including: parsers and serializers

RDFLib 1.8k Jan 02, 2023
An a simple sistem code in python

AMS OS An a simple code in python ⁕¿What is AMS OS? AMS OS is an a simple sistem code writed in python. This code helps you with the cotidian task, yo

1 Nov 10, 2021
Python script for diving image data to train test and val

dataset-division-to-train-val-test-python python script for dividing image data to train test and val If you have an image dataset in the following st

Muhammad Zeeshan 1 Nov 14, 2022
This repository contains each day of Advent of Code 2021 that I've done.

Advent of Code - 2021 I will use this repository as my Advent of Code1 (AoC) repo for the 2021 challenge. I'm changing how I am tackling the problems

Brett Chapin 2 Jan 12, 2022
CupScript is a simple programing language made with python

CupScript CupScript is a simple programming language made with python It includes some basic functions, variables, loops, and some other built in func

FUSEN 23 Dec 29, 2022
A subleq VM/interpreter created by me for no reason

What is Dumbleq? Dumbleq is a dumb Subleq VM/interpreter implementation created by me for absolutely no reason at all. What is Subleq? If you haven't

Phu Minh 2 Nov 13, 2022
Structured, dependable legos for starknet development.

Structured, dependable legos for starknet development.

Alucard 127 Nov 23, 2022
Python implementation of the Lox language from Robert Nystrom's Crafting Interpreters

pylox Python implementation of the Lox language from Robert Nystrom's Crafting Interpreters. https://craftinginterpreters.com. This only implements th

David Beazley 37 Dec 28, 2022
ERPNext Easy Letterhead

ERPNext Easy Letterhead Intro Quality letterheads are a problem for non-technical users. So we've built (really hacked together) a slightly easier sol

Bantoo 3 Jan 02, 2023