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

Neural Codes for Image Retrieval #14

Open
chullhwan-song opened this issue Jul 9, 2018 · 1 comment
Open

Neural Codes for Image Retrieval #14

chullhwan-song opened this issue Jul 9, 2018 · 1 comment

Comments

@chullhwan-song
Copy link
Owner

https://arxiv.org/abs/1404.1777

@chullhwan-song
Copy link
Owner Author

chullhwan-song commented Jul 9, 2018

what?

  • neural code
  • neural code만의 새로운 landmark 데이터셋 공개
    • we chose a (semi)-automated approach for that.
    • by selecting 10,000 most viewed landmark Wikipedia pages (over the last month).
    • we used the title of the page as a query to Yandex image search engine.
    • then downloaded 1000 top images returned in response to the query (or less, if the query returned less images).
    • 한마디로, text기반의 검색으로 이미지를 수집
    • 따라서, 노이즈가 좀 많음. > triplet r-mac(DIR)에서는 이런 데이터 특성으로 인해 noisy를 제거하여 학습셋으로 사용. 덧붙혀, google delf 도 이를 이용
    • 최종적으로, 672 classes and 213,678 images.
  • imagenet 기반의 학습 모델을 이용. 그리고 위의 landmark set을 학습셋으로한 fine-tuning한 deep feature를 적용된 초기 논문
    image
    • alexnet > "ImageNet Classification with Deep Convolutional Neural Networks"
    • deep feature : fully-connected layers 를 이용한다고 보이지만, 위의 그림에서는 conv feature map도 적용한 모습이 보임
  • 최종적으로 PCA를 적용한 feature

실험

  • full descriptor
    image
  • compression
    image
  • imagenet model vs fine-tuning ?
    image

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

No branches or pull requests

1 participant