Search similar images.The similarity between images are modeled by CNN (Convolutional Neural Networks) model.Real-time search by K-D Tree.
Please refer to code/readme.txt for running
Visulaization of Images
Images are plotted by feature values
Search System (CNN model + K-D tree)
Input image and return top-k similar images
Image similarity by features of CNN model
Each image is represented by feature values of the-sencond-to-last FC layer
(the last FC layer outputs labels)
Index by K-D tree
Each image is indexed by its feature values by K-D tree.
Build index: n*log(n)
CNN Models Optimization
Real-time Search by K-D tree
Not matched case
To search for more types of images, we should use more different and new images to train models.
Subscribe to Python Awesome
Get the latest posts delivered right to your inbox