Boosted translation-tolerable classifiers for fast object detection

► We define the Maximal Translation-Tolerable Region (MTTR) to measure classifiers' sensitivity to translation-variance. ► We train a cascaded classifier that the MTTRs of the earlier strong classifiers in it are large and vice versa. ► We propose a Granularity-Adaptively-Tunable (GAT) search according to the strong classifiers' MTTRs. ► The recall rate is probably approximately admissible in the GAT search.