ConvNet, Visualizing and Understanding
2018-11-15
CNN을 내부영역을 시각화(Visualize)하는 방법과 Image를 Reconstruction하는 아이디어
와 이를 구현하기 위한 방법
을 알아보겠습니다.
목차
- Visualize
- First Layer: Visualize Filters(kernels)
- Visualizing Activations
- Maximally Activating Patches
- Last Layer: Nearest Neighbors
- Last Layer: Dimensionality Reduction
- Occlusion Experiments
- Saliency Maps
- Saliency Maps: Segmentation without supervision
- Generate
- Fooling Images / Adversarial Examples
- DeepDream: Amplify existing features
- Feature Inversion(Reconstruct)
- Texture
- Texture Synthesis
- Texture Synthesis: Nearest Neighbor
- Neural Texture Synthesis: Gram Matrix
- Neural Texture Synthesis
- Neural Texture Synthesis: Texture = Artwork
- Style
- Neural Style Transfer: Feature + Gram Reconstruction
- Neural Style Transfer
- Neural Style Transfer: Multiple Style Images
- Fast Style Transfer
- One Network, Many Styles
표
마크다운으로 표 만들어주는 사이트(markdown table generator)
Reference
- https://junjiwon1031.github.io/2017/09/08/Single-Shot-Multibox-Detector.html
- https://sites.google.com/site/bimprinciple/in-the-news/dibleoning-euliyonghangaegchegeomchulr-cnnyolossd
- https://m.blog.naver.com/PostView.nhn?blogId=sogangori&logNo=221007697796&proxyReferer=https%3A%2F%2Fwww.google.co.kr%2F
- https://m.blog.naver.com/sogangori/221009464294