This paper was accepted on the CV4Metaverse Workshop at CVPR 2025.
With speedy developments in digital actuality (VR) headsets, successfully measuring stereoscopic high quality of expertise (SQoE) has develop into important for delivering immersive and comfy 3D experiences. Nevertheless, most present stereo metrics concentrate on remoted features of the viewing expertise resembling visible discomfort or picture high quality, and have historically confronted information limitations. To handle these gaps, we current SCOPE (Stereoscopic COntent Desire Analysis), a brand new dataset comprised of actual and artificial stereoscopic pictures that includes a variety of widespread perceptual distortions and artifacts. The dataset is labeled with choice annotations collected on a VR headset, with our findings indicating a notable diploma of consistency in person preferences throughout completely different headsets. Moreover, we current iSQoE, a brand new mannequin for stereo high quality of expertise evaluation skilled on our dataset. We present that iSQoE aligns higher with human preferences than present strategies when evaluating mono-to-stereo conversion strategies.
- †Work finished whereas at Apple
- ‡ Weizmann Institute of Science
- § Massachusetts Institute of Expertise
- ¶ Impartial researcher
