In this Repository, I generated models that can predict real images that are trained on the computer-generated image of fashion items. I also create embeddings to visualize data in 3-dimensions with PCA and T-SNE.
In this file, I preprocessed data and generate a convolutional neural network with a saving model. In the section of Import_Resize_Reshape_prediction, I did all the transformation of images without saving images as a file without image generator function so, everyone can understand how the whole process works of image transformation. In the end, I predicted 30 images that I downloaded from the internet.
In this file, I generated embeddings from the test dataset. by embeddings, we can use tensroboard to see data in 3-dimensions with help of PCA(Principal component analysis) and T-SNE(t-distributed stochastic neighbor embedding) which help us to understand how data works in multi-dimension. with T-SNE we can also understand which fashion items have similar features that define how our model can predict the wrong prediction between two fashion items.
If you want to generate Visualization, you need to follow the steps given in Report.pdf because due to the larger size, I can not import all the files.