Super Easy Way of Building Image Search with Keras
Download as PPTX, PDF1 like330 views
The document discusses Hiroki Tanioka's work on building an image search system using Keras, with a focus on deep learning for information retrieval. It highlights Tanioka's background in research and his current interests in analyzing sports and medical data. The content includes references to notable concepts and tools such as Lucene, deep learning, and k-nearest neighbors, along with links to relevant resources.
Super Easy Way of Building Image Search with Keras
- 1. Super Easy Way of Building Image Search with Keras Introduction of Deep Learning for Information Retrieval Researchers in LIARR 2017. (11, Aug. 2017) Hiroki Tanioka
- 2. Hiroki Tanioka Research Associate Ph.D. Tokushima University Center for Administration of Information Technology I am working at a university in Japan. After 18 years of researcher and engineer in some software companies. My research area is Information Retrieval and Machine Learning. These days, I am interested in analyzing sports data and medical data.
- 5. How to search by Deep Learning?
- 6. What is features for Search Engine? Synset features
- 7. What is difference to former product? http://www.lire-project.net/ http://scikit-learn.org/ + (k-nearest neighbor)(color and texture characteristics) (Deep Learning) (Inverted Index)
- 9. Next challenge is using Deep Features! http://cvlab.postech.ac.kr/research/deconvnet/ Deep CNN Deep Features
Editor's Notes
- #3: Shuji Nakamura is graduated from Tokushima University, who is one of Blue LED Nobel Prize inventers in 2014.
- #4: Lucene is a powerful search core engine, you know. Solr and Elasticsearch include Lucene as search engine. I will use elasticsearch because It has RESTful API.
- #5: Deep Learning is a powerful algorithm for image recognition. Especially, Deep Convolutional Networks works well for object recognition in images.
- #6: I am trying simple image search with Keras and Elasticsearch. I got Dog vs Cat images from Kaggle. Kaggle has many data collections for data science and machine learning.
- #7: Keras has VGG16 model which is pretrained. VGG16 model can output synset features. Of course, you can use deep features
- #8: Keras has VGG16 model which is pretrained. VGG16 model can output synset features. Of course, you can use deep features
- #9: Please check the github site.
- #10: Of course, you can use deep features