Reproduce “Can We Get Rid of Handcrafted Feature Extractors? SparseViT: Nonsemantics-Centered, Parameter-Efficient Image Manipulation Localization through Spare-Coding Transformer” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Lei Su, Xiaochen Ma, Xuekang Zhu, Chaoqun Niu, Zeyu Lei, Ji-Zhe Zhou
SparseViT localizes manipulated pixels by limiting which separated patches exchange information, then fusing multi-scale features into a pixel-level mask, without a separate forensic feature extractor. On four benchmark datasets, it achieves mean pixel-level F1 of 0.671 versus 0.627 for TruFor. At 512×512, it uses 46.2 GFLOPs and 50.3M parameters; these results do not establish real-world generalization or hardware speedup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.