Publications

The paper addresses the challenge of quality assessment for 360-degree images by introducing LGT360IQ, a dual-branch framework that effectively combines local and global information for improved accuracy. Their extensive evaluation on multiple datasets highlights the framework's superior performance compared to traditional methods.

Our study introduces ST360IQ, a no-reference quality assessment method that extracts and evaluates tangent viewports from key areas of the image using vision-transformer-based modules. This approach aggregates scores to produce a final quality assessment. Tested on OIQA and CVIQ datasets, ST360IQ shows improved correlation with human-perceived image quality compared to current methods.