We propose a novel unsupervised approach based on a two-stage object-centric\nadversarial framework that only needs object regions for detecting frame-level\nlocal anomalies in videos. The first stage consists in learning the\ncorrespondence between the current appearance and past gradient images of\nobjects in scenes deemed normal, allowing us to either generate the past\ngradient from current appearance or the reverse. The second stage extracts the\npartial reconstruction errors between real and generated images (appearance and\npast gradient) with normal object behaviour, and trains a discriminator in an\nadversarial fashion. In inference mode, we employ the trained image generators\nwith the adversarially learned binary classifier for outputting region-level\nanomaly detection scores. We tested our method on four public benchmarks, UMN,\nUCSD, Avenue and ShanghaiTech and our proposed object-centric adversarial\napproach yields competitive or even superior results compared to\nstate-of-the-art methods.\n