GithubHelp home page GithubHelp logo

ljl02521 / 2018-bd-xjd-2- Goto Github PK

View Code? Open in Web Editor NEW

This project forked from maozezhong/2018-bd-xjd-2-

0.0 1.0 0.0 19 KB

2018百度西交大大数据竞赛-商家招牌的分类与检测-复赛(23/1139)

Python 100.00%

2018-bd-xjd-2-'s Introduction

比赛介绍

  • 针对检测+分类任务,我们提供9000张带有位置信息和类别信息的图像数据用于训练,4351张图像用于评估测试。该数据集全部来源于百度地图淘金,选取了60类常见品牌类别。比如,肯德基,星巴克,耐克等。

比赛历程

  • 7月13号结束,7月初开始做.中间尝试了
  • 最后还是tow-stage的faster-rcnn正确度高(至少在我的实验中是这样,当然时间有限,设备有限单卡1080,实验结论不完备)
  • 数据处理方面 : 使用了针对检测的数据增强,包括旋转,平移,加噪,改亮度,具体实现见DataAugmentForObejctDetection.py这个脚本
  • trick方面 : 1)softnms, 2)模型融合(具体见merge_box中的脚本)
  • batchsize基本上是1,设备受限上不去了; lr初始一般设的0.001, 每5轮降为原来的十分之一; 输入尺度试过600和800
  • 最后线上为0.8576,排名23,没苟进决赛,哎...

脚本说明:

  • merge_box:
    • csv_2_txt_for_merge.py : 根据结果csv产生中间txt文件
    • merge_res.py : 融合并产生最终csv结果文件
  • show_boundingbox_on_pic.py : 可视化脚本
  • DataAugmentForObejctDetection.py : 数据增强脚本
  • densenet.py : pytorch, 基于densenet backbone的faster rcnn模型结构(未实验), 参考vision/torchvision/models/densenet.py
  • resnext.py : pytorch, 基于resnext backbone的faster rcnn模型结构, 参考ResNeXt-PyTorch/resnext.py, 其实就是在resnet的基础上加了多通道并行.
  • nms.py : 常规nms,以及softnms

学习姿势

  • faster多尺度训练,多尺度预测,tesnsorlayer数据增强,结果ensemble可以到89
  • 基于fpn的faster-rcnn
  • detectron单模型可以到89
  • predict的时候augment, detectron中的config文件下有例子
  • 调参方法,何凯明论文,detctron有论文链接
  • 增强后的数据作为验证集,把60类验证集AP保存,取每一类最好的ap的模型进行集成
  • 数据增强库 : imgaug, emmm...应该比我自己整的靠谱点
  • 调整分类和bbox的loss权重
  • 使用sniper模型
  • ssd上89!!!但是没说用了啥技巧....

所有代码链接

to do list

2018-bd-xjd-2-'s People

Contributors

maozezhong0 avatar maozezhong avatar

Watchers

James Cloos avatar

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. 📊📈🎉

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google ❤️ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.