Doubts regarding pop signals #6
Description
I read the POP model paper and it seems very interesting and promising. I am trying to generate pop signals for my dataset but I have trouble understanding few things. From what I understand:
[Please correct me if I am wrong]
Images are extracted from google and refined using CL. These refined images are inferred through a 'style popularity' prediction model based on Fashion Forward dataset to predict 'POP style signals' for the particular year and the model is based on ResNet, detects features of the image and predicts it's popularity. The popularity is predicted for images in the VISUELLE dataset, representing normalised values for the popularity and stored as pop.pt file.
- Are there any code references I can refer to generate the POP style signals from images?
- Do 52 values in the pop.pt file for each item represent one year of fashion trend for the corresponding dress in the image?
Really appreciate the efforts gone into creating the model and the paper!