This project allows images to be automatically grouped into like clusters using a combination of machine learning techniques.
Unsupervised machine learning is a technique that can used to automatically classify or group data together that has no preidentified structure (as opposed to supervised learning where an "expert" has labeled a training data set).
There are two steps involved in the process. The first is to use a pre-trained deep learning model to extract a feature vector of each image in the collection. Once we have the vector (which is an array of floating point values) it is then be passed into a t-SNE function, which takes all of the arrays and reduces them down to two values: X and Y. These two values can then be plotted against each other to produce a graph which Zegami can use as a filter.
Uses one of the pre-trained deep learning models avaliable in Keras to extract a feature vector for all images in a source directory.
I used the following guide to install Keras with TensorFlow using conda.
The script expects as an argument the path to a tab separated file that has at a minimum a column called 'id' and another called 'image' which contains the file name. The images need to be located in a directory called images which is located in the same directory as the source file.
For example if a file called example.tsv contains a single record:
| id | image |
|---|---|
| 1 | 1.jpg |
Then it would have the following directory structure:
.
+-- example.tsv
+-- images
| +-- 1.jpg
The results are saved to a tab separated file postfixed with '_features.
Takes a comma separated list of values and runs them through a t-SNE function. The result is then saved back to a tab separated file postfixed with '_tsne'.