GIFTBench: Diagnosing Generalization in Image Forgery Localization and Informing Model Design

Reliable evaluation of image forgery localization (IFL) requires assessing mod- els under diverse distribution changes, yet existing benchmarks often cover lim- ited manipulation conditions or entangle multiple factors in cross-dataset eval- uation. Consequently, aggregate performance provides an incomplete view of localization generalization. We introduce GIFTBench, a multi-axis benchmark of 115,013 manipulated images with pixel-level annotations spanning manipu- lation source, semantic target, editing operation, and composition complexity. GIFTBench supports axis-specific transfer analysis and evaluation on twelve ex- ternal datasets. Its diagnostic studies reveal asymmetric cross-source transfer, recall-dominated failures, and heterogeneous degradation across semantic, oper- ational, and compositional changes. Beyond diagnosis, the scale and diversity of GIFTBench provide a substantially broader training distribution than conven- tional IFL datasets. Training representative localizers on GIFTBench consistently improves their aggregate transfer to external datasets, showing that the benchmark serves not only as an evaluation tool but also as an effective training resource for cross-domain localization. Guided by the diagnostic findings, we further develop FORENSCOPE, a detection and localization framework combining classification- adapted representations with multi-depth, multi-scale spatial features, learned layer fusion, and selective coarse-scale conditioning. Experiments show im- proved cross-dataset localization while retaining image-level detection capability. The GIFTBench dataset resources and ForenScope code are available at https: //anonymous.4open.science/r/GIFTBench-ForenScope.