A Multilevel Visual and Textual Framework for Near-Duplicate Diagram Detection in Electronic Documents
A cascaded multimodal framework combining perceptual hashing, Siamese Residual Network with 18 layers, Siamese Residual Network with 18 layers, Distillation with No Labels, ver. 2 (DINOv2)visual representations, and a text-similarity classifier is proposed to detect near-duplicate diagram detection in electronic docume...