Pre-trained ResNet50 Model for Website Screenshot Overlay Detection
1 min read
This post provides access to the pre-trained deep learning model that achieved the highest accuracy in our study on detecting overlays (pop-ups, banners, etc.) in website screenshots. We compared three CNN architectures on a dataset of 1,397 labeled screenshots, and the ResNet50 model, utilizing transfer learning from ImageNet weights, demonstrated superior performance, achieving 96% accuracy on our test set. This model takes a 224x224 website screenshot as input and outputs a binary classification indicating the presence or absence of an overlay. It can be integrated into automated web analysis workflows to identify and potentially handle overlay interference. More details are available in our ICAR'15 paper. Download the trained model via the link below.
git clone https://hf.co/goker/overlay