Build a convolutional neural net for image similarity -- 2

Build two statistical models: Model 1) Input: one query image, a dataset of product images; Output: visually similar images from product image dataset matching query image, together with a similarity score for each Model 2) Input: one set of images; Output: a clustering of the input images Idea: since Model 1 can compute image similarity between images, Model 2 can simply run Model 1 on all pairs of images to get similarities, then use a standard clustering algorithm based on the similarity matrix. Model 2: Given a function model1(X) -> {a:0.9, b:0.7, c:0.6, d:0.4}, Gabi will build model 2. Accuracy / Precision / Recall will be computed on the top 10 results, and we desire 80% accuracy on those. Datasets to be provided: Dataset 1: Lifesytle images - aprox 4.5k images of fashion related lifestyle images. Provided in a zip folder with just the individual images. [login to view URL] Dataset 2: Products: aprox 50k product file with, ID, Name, Description, Main Image URL and Alternative Image URL. [login to view URL] Dataset 3: Visenze results for all images and associated product results with their similarity score. [login to view URL]