Core Concepts
Proposing a novel contrastive pre-training framework tailored for PCQA to enhance quality-aware representations and improve NR-PCQA methods.
Abstract
The article introduces a novel contrastive pre-training framework, CoPA, designed for Point Cloud Quality Assessment (PCQA). It addresses the challenge of limited labeled data in learning-based NR-PCQA methods. CoPA generates anchors through local patch mixing to preserve distortion patterns and employs content-wise and distortion-wise contrasts in pre-training. In the fine-tuning stage, a semantic-guided multi-view fusion module integrates features from different perspectives. Experimental results show superior performance compared to state-of-the-art methods on popular benchmarks LS-PCQA, SJTU-PCQA, and WPC datasets. The method demonstrates robustness across datasets and outperforms other NR-PCQA models with less labeled data.
Stats
No key metrics or figures provided in the content.
Quotes
"Extensive experiments show that our method outperforms the state-of-the-art PCQA methods on popular benchmarks."
"To tackle the challenge of label scarcity, we propose a contrastive pre-training framework tailored for PCQA."
"Our model presents competitive and generalizable performance compared to the state-of-the-art NR-PCQA methods."