Based on the use of the Fashion 144K dataset in the papers: Joint Ranking and Classification using Weak Data for Feature Extraction(http://www.f.waseda.jp/hfs/SimoSerraCVPR2016.pdf) and An Analysis of Human-centered Geolocation(https://arxiv.org/abs/1707.02905).
Predicts the Fashionability score (1-10) of images based on the Fashion 144k Dataset.
I used Resnet50 for Feature extraction process connecting to a 10-d fc layer. The model was trained on a subset of the dataset, that is, 3000 train mages and 1000 validation images.
Python, Keras, Hyperas, Sklearn, Numpy, Matplotlib
After training longer, my accuracy came out to be around 14% compared to 17% in the original paper. Hyperparameters optimized using Hyperas.