**Advancing Zero-Shot Digital Human Quality Assessment through Text-Prompted Evaluation** (2023)
The work stems from the challenges of collecting 3D digital humans, which is more difficult and expensive than 2D images/videos, leading to a scarcity of subjective databases for 3D digital humans. In response, the researchers have created the first perceptual quality assessment database for full-body digital humans, SJTU-H3D. Unlike many large-scale image/video quality assessment databases that enable data-driven methodologies, the limited number of DHQA databases can introduce bias in supervised methods, affecting their generalization ability. To overcome this, the researchers propose a zero-shot, no-reference quality assessment method.
=> https://arxiv.org/abs/2307.02808v1