Gradio

Gazer 1.0: Ad Attention Prediction

This app accompanies: "Contextual Advertising with Theory-Informed Machine Learning", manuscript submitted to the Journal of Marketing.
App Version: 1.0, Date: 10/24/2024.
Note: Gazer 1.0 does not yet include LLM generated ad topics. Future updates will include this in a GPU environment.

If you only have a pdf image file, first convert it here to png file and download:

{}

Progress

Instructions:

The screen size should remain the same during processing.
Click to upload or drag the entire image (jpg/jpeg/png file) that contains BOTH ad and its context;
Draw bounding boxes in the order of: (each element can have more than 1 boxes; remember the number of boxes for each element you draw)
   (a) Brand element(s) (skip if N.A.)
   (b) Pictorial element(s), e.g. Objects, Person etc (skip if N.A.)
   (c) Text element(s) (skip if N.A.)
   (d) The advertisement.
Put in number of bounding boxes for each element, product category, ad location and attention type.

NOTE: ResNet50 Heatmap could take around 20-80 seconds under current CPU environment.

Two example ads are avialable for download:

{}

Number of brand bounding boxes drawn

Number of pictorial bounding boxes drawn

Number of text bounding boxes drawn

Check if you draw a bounding box for the entire ad (Note: this is a must-do)

Product Category

Ad Location

Gaze Type

Predicted Gaze (sec). If you see a value of -1, it means no ad bounding box is drawn!!

ResNet50 Heatmap (Hotter/Redder regions show more pixel contribution.)

Disclaimer: This app is provided for free and for academic use only. The authors take no responsibility for your use of the information contained in or linked from these web pages.