研究目的
To estimate the perceived visual quality of 3D meshes without having access to the reference using a convolutional neural network (CNN) framework.
研究成果
The proposed CNN-based method successfully predicts the visual quality of distorted meshes with high correlation to human judgment, without requiring the reference mesh. This makes it suitable for practical applications where the reference is not available.
研究不足
The method's performance is dependent on the accuracy of the saliency computation and the selection of patches. The generalization to other types of distortions or meshes not included in the training set is not discussed.
1：Experimental Design and Method Selection:
The proposed CNN architecture is designed to estimate the visual quality of 3D meshes by processing small patches selected based on their saliency level.
2：Sample Selection and Data Sources:
Two datasets of distorted and scored meshes are used for evaluation.
3：List of Experimental Equipment and Materials:
Not explicitly mentioned.
4：Experimental Procedures and Operational Workflow:
The process involves computing the visual saliency of the 3D mesh, rendering 2D projections, splitting these projections into small patches, selecting patches based on saliency, and feeding them into the CNN for quality estimation.
5：Data Analysis Methods:
The performance is evaluated using Pearson linear correlation coefficient (rp) and Spearman rank-order correlation coefficient (rs).
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
