Measure the detector on a labelled dataset
Select a benchmark set (CASIA v2, CoMoFoD, Columbia, MICC-F220 or your own collection). Each image is scored in your browser and compared with its known category, producing the accuracy, precision, recall, F1, ROC-AUC and confusion matrix your write-up needs.
1. Label your images
The known category is read from the folder or file name. Use folders named authentic, copy-move, spliced and ai-generated, or prefix the file names the same way (for example Sp_001.jpg, Au_014.jpg).
Every image runs the full forensic suite, so large sets take a while — expect a few seconds per photo. Results appear as they are produced.