Caught on Camera (CoC) Dataset: On-Screen Deepfakes Recaptured with Mobile and Wearable Cameras

Caught on Camera (CoC) Dataset Dataset Summary Caught on Camera (CoC) accompanies the paper Caught on Camera: Toward Evaluating and Defending On-Screen Deepfakes With Mobile and Wearable Camera Devices, to appear at RAID 2026. The dataset supports the study of deepfake detection when media displayed on physical screens is recaptured using smartphones or smart glasses. It contains 29,900 aligned real/fake sample pairs collected under 103 display-recapture configurations, including 14,950 real and 14,950 fake samples. Dataset Composition The 103 configurations comprise: 19 controlled single-factor configurations, examining capture distance, viewing angle, ambient lighting, and Moiré interference 84 multi-factor configurations representing natural combinations of display, camera, geometry, lighting, reflection, and device-side image processing Capture Devices and Displays Capture devices: iPhone 17 Pro Vivo V21 Meta Wayfarer Rokid glasses Display devices: Redmi display AOC display MAXHUB 55-inch display MAXHUB 86-inch display STARZEN display SEEWO display