Crossscore: towards multi-view image evaluation and scoring

Conference item

We introduce a novel cross-reference image quality assessment method that effectively fills the gap in the image assessment landscape, complementing the array of established evaluation schemes – ranging from full-reference metrics like SSIM [59], no-reference metrics such as NIQE [32], to general-reference metrics including FID [17], and Multi-modal-reference metrics, e.g. CLIPScore [16]. Utilising a neural network with the cross-attention mechanism and a unique data collection pipeline from NVS optimisation, our method enables accurate image quality assessment without requiring ground truth references. By comparing a query image against multiple views of the same scene, our method addresses the limitations of existing metrics in novel view synthesis (NVS) and similar tasks where direct reference images are unavailable. Experimental results show that our method is closely correlated to the full-reference metric SSIM, while not requiring ground truth references.

Authors: Wang, Z; Bian, W; Prisacariu, VA

• Peer review status: Peer reviewed
• Publisher: Springer
• Host title: Computer Vision – ECCV 2024: 18th European Conference, Milan, Italy, September 29–October 4, 2024, Proceedings, Part IX
• Pages: 492-510
• Series: Lecture Notes in Computer Science
• Series number: 15067
• Place of publication: Cham, Switzerland
• Publication date: 2024-10-22
• Acceptance date: 2024-07-01
• Event title: 18th European Conference on Computer Vision (ECCV 2024)
• Event location: Milan, Italy
• Event website: https://eccv.ecva.net/Conferences/2024
• Event start date: 2024-09-29
• Event end date: 2024-10-04
• DOI: 10.1007/978-3-031-72673-6_27
• EISSN: 1611-3349
• ISSN: 0302-9743
• EISBN: 9783031726736
• ISBN: 9783031726729
• Language: English
• Keywords: cross-reference image assessment; novel view synthesis