ImageNet Roulette is a computer vision research project that uses a large-scale image database to explore the effects of random perturbations on objects and scenes. The project is divided into two phases: the first phase involves perturbing images to create a large number of variants, while the second phase uses a rolling window of variants to study the effects of varying the perturbation intensity, duration, and location.
ImageNet Roulette has already produced some interesting findings. For example, it has shown that increasing the intensity of perturbations can lead to more interesting and diverse object shapes, while also reducing the overall object quality.
Additionally, it has also shown that increasing the duration of perturbations can lead to more natural-looking object color distributions. However, further work is required in order to better understand the effects of varying perturbation intensity, duration, and location on object and scene quality.
