Hello, author. Your work is fantastic, but I have some questions regarding your paper. In the paper, the description about the sources of multimodal knowledge, '... textual specifications such as structural descriptions, tolerance constraints, and surface requirements, and Eprior denotes expert knowledge about plausible defect modes and critical regions. These complementary cues jointly inform what kind of anomaly should be synthesized, where it should be placed, and how strong the geometric perturbation should be.' For this part, I would like to know how you obtained this knowledge. For example, with the Real3D-AD dataset you used, when synthesizing anomalies for this dataset, observing the visualizations in Fig.4 and Fig.5, the paper states, 'Real anomalies are shown as references, and MPAS is then applied to synthesize the same defect types under comparable geometric conditions;' can we understand this as you having manually collected information about the real anomalies in the Real3D-AD dataset and then using this knowledge to synthesize anomalies?
Hello, author. Your work is fantastic, but I have some questions regarding your paper. In the paper, the description about the sources of multimodal knowledge, '... textual specifications such as structural descriptions, tolerance constraints, and surface requirements, and Eprior denotes expert knowledge about plausible defect modes and critical regions. These complementary cues jointly inform what kind of anomaly should be synthesized, where it should be placed, and how strong the geometric perturbation should be.' For this part, I would like to know how you obtained this knowledge. For example, with the Real3D-AD dataset you used, when synthesizing anomalies for this dataset, observing the visualizations in Fig.4 and Fig.5, the paper states, 'Real anomalies are shown as references, and MPAS is then applied to synthesize the same defect types under comparable geometric conditions;' can we understand this as you having manually collected information about the real anomalies in the Real3D-AD dataset and then using this knowledge to synthesize anomalies?