24
https://pubmed.ncbi.nlm.nih.gov/38091756
The abstract describes a new method for creating adversarial attacks on video classification models, called DeepSAVA, which uses a sparse perturbation strategy and structural similarity index to maintain human imperceptibility while achieving high attack success rate and adversarial transferability. The method can also be used to improve the robustness of video classification models through adversarial training.