Real-world datasets follow an imbalanced distribution, which posessignificant challenges in rare-category object detection. Recent studies tacklethis problem by developing re-weighting and re-sampling methods, that utilisethe class frequencies of the dataset. However, these techniques focus solely onthe frequency statistics and ignore the distribution of the classes in imagespace, missing important information. In contrast to them, we propose FRActalCALibration (FRACAL): a novel post-calibration method for long-tailed objectdetection. FRACAL devises a logit adjustment method that utilises the fractaldimension to estimate how uniformly classes are distributed in image space.During inference, it uses the fractal dimension to inversely downweight theprobabilities of uniformly spaced class predictions achieving balance in twoaxes: between frequent and rare categories, and between uniformly spaced andsparsely spaced classes. FRACAL is a post-processing method and it does notrequire any training, also it can be combined with many off-the-shelf modelssuch as one-stage sigmoid detectors and two-stage instance segmentation models.FRACAL boosts the rare class performance by up to 8.6% and surpasses allprevious methods on LVIS dataset, while showing good generalisation to otherdatasets such as COCO, V3Det and OpenImages. We provide the code athttps://github.com/kostas1515/FRACAL.