Touch vs. No-Touch Video Dataset from a See-Through Perspective

This dataset contains 16,841 color images (448×448 pixels, RGB) of a stylus in contact or not in contact with a surface, collected from a see-through optical setup. Each image is paired with a binary label: 1 for contact and 0 for no-contact. The dataset was collected as part of PhD research at Massey University, focusing on human-computer interaction and stylus detection in transparent interaction surfaces. The images were captured using a synchronized high-speed camera and manually labeled based on contact state. Data is stored in a TFRecord format (all_data.tfrecord), with two datasets: /images (shape: N×448×448×3, dtype: uint8, RGB) and /labels (shape: N, dtype: bool). This dataset is intended for use in computer vision, interaction design, and tactile sensing research.