Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Vehicle Detection in Aerial Images Based on Region Convolutional Neural Networks and Hard Negative Example Mining #10

Open
KenichiSasaki opened this issue Oct 23, 2019 · 0 comments
Labels
CNN Feature Engineering Feature Extraction Machine Learning Method Employ machine learning method Object Detection

Comments

@KenichiSasaki
Copy link
Collaborator

概要

  • アルゴリズム系論文(2017)
  • 車に似た形状を持つfeature(建物etc)を候補から削除し精度を向上するHard Negative Example Miningを提案
    • Faster R-CNNにアーキテクチャを基にネットワーク構築

アルゴリズム

  • 元画像から切り取り,回転等のData augmentation
  • Faster R-CNNに用いられているRegional Proposal Network (RPN)を用いて車の候補を抽出
  • 決定木に基づくBoostingアルゴリズムで正例と負例を区別
    2019-10-23_15h06_39

実装

  • Munich vehicle datasetを利用
  • 既存のアルゴリズムより精度が少し向上
  • 木に隠れている車や隣接する車の識別は失敗が多く見られた

所感

  • 元画像とアノテーションデータとの誤差に着目して精度向上するのは面白い
  • トレーニングデータから最大限特徴を学習する手順としてはFeature engineering的アプローチ
  • End to endでできたら車以外にも応用可能?
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
CNN Feature Engineering Feature Extraction Machine Learning Method Employ machine learning method Object Detection
Projects
None yet
Development

No branches or pull requests

1 participant