This started as an actual experiment I decided to make a while ago. The experiment was to check how much a well-trained model could hold its output by slightly changing the input image. I'll call this changes RISPP - Random Incremental Single Pixel Perturbations.
It's everything on the ispp.ipynb notebook. Run pip3 install -r requirements.txt to install needed packages.
You could use your own model and images. Check it out!
This is already a well-documented field, that started around 2015 and still growing stronger since then. The research in this field keeps pushing the models to their limits, both for attack or/and defense. For additional info, check those papers that were also used to fundament this experiment: Haotian Zhang, XuMa (2022), Hong Wang et al. (2021), Atiye Sadat Hashemi & Saeed Mozaffari (2021), Kevin Eykholt et al. (2018), Xiaoshuang Shi et al.(2022), Jiawei Su et. al (2017).