OcuBench
A multi-source benchmark dataset designed for the eyeglass reflection removal task. It includes synthetic image pairs, real pseudo-paired data, and real-world test images, supporting systematic evaluation of model generalization capability across complex, realistic scenarios.
Timeline (last 90 days)
A study published on arXiv proposes OcuBench, a multi-source benchmark dataset, and OcuFlow, an eye-adaptive pixel-level MeanFlow framework, for addressing the problem of eyeglass reflection removal.
OcuBench contains 10,280 controllably synthesized image pairs for paired evaluation under controlled conditions.
OcuBench contains 732 real-input pseudo-paired samples.
OcuBench contains 458 independent real-world test images.
The research team stated that the code and datasets for both OcuBench and OcuFlow will be publicly released.
All Facts (5)
OcuBench contains 10,280 controllably synthesized image pairs for paired evaluation under controlled conditions.
50%UnverifiedOcuBench contains 732 real-input pseudo-paired samples.
50%UnverifiedOcuBench contains 458 independent real-world test images.
50%UnverifiedThe research team stated that the code and datasets for both OcuBench and OcuFlow will be publicly released.
50%UnverifiedA study published on arXiv proposes OcuBench, a multi-source benchmark dataset, and OcuFlow, an eye-adaptive pixel-level MeanFlow framework, for addressing the problem of eyeglass reflection removal.
50%