This is a pytorch project used for detecting backdoors at three different levels.
For the model level, we select ABS and Neural Cleanse. For the dataset level, SCAn, Spectral Signatures, and SPECTRE are chosen. And for the input level, we use STRIP and Denoising AE.