ARIQA-3DS: A Stereoscopic Image Quality Assessment Dataset for Realistic Augmented Reality

A team of researchers has released ARIQA-3DS, the first large-scale stereoscopic dataset designed to measure image quality and user experience in Augmented Reality. The dataset, created by Aymen Sekhri, Seyed Ali Amirshahi, and Mohamed-Chaker Larabi, addresses a critical gap in AR development: the lack of realistic, ecologically valid data to train quality assessment models. Existing datasets often fail to capture the complex 'visual confusion' between real and virtual layers, relying on monocular views or simplified backgrounds. ARIQA-3DS solves this by fusing high-resolution, stereoscopic 360-degree captures of real-world environments with diverse virtual foregrounds, all under controlled transparency and degradation conditions.

Comprising 1,200 AR viewports, the dataset is the result of a comprehensive subjective study involving 36 participants using a video see-through head-mounted display. The study collected not only quality ratings but also simulator-sickness indicators, providing a holistic view of user comfort. Key findings reveal that perceived quality is primarily driven by the degradations in the virtual foreground and is significantly modulated by its transparency level. Importantly, the study found that oculomotor and disorientation symptoms increased progressively during viewing but remained manageable, offering crucial data for comfort thresholds. By publicly releasing ARIQA-3DS, the researchers aim to establish a definitive benchmark for developing the next generation of AR quality assessment algorithms, which are essential for creating immersive and comfortable consumer AR experiences.