Improving Object Localization with Fitness NMS and Bounded IoU Loss

We demonstrate that modern detection methods are designed to identify only a sufficently accurate bounding box, rather than the best available one. To address this issue we propose a simple and fast modification to the existing methods called Fitness NMS. This method is tested with the DeNet model and obtains a significantly improved MAP at greater localization accuracies without a loss in evaluation rate. We also derive a novel bounding box regression loss based on a set of IoU upper bounds and investigate simple RoI clustering schemes for improving evaluation rates for the DeNet wide model variants. Following these novelties we provide an analysis of input image dimensions with the DeNet model. We obtain a MAP[0.5:0.95] of 33.6%@79Hz and 41.8%@5Hz for MSCOCO and a Titan X (Maxwell). These results provide further evidence that DeNet and Fitness NMS methods can be scaled to provide highly competitive performance at a broad range of evaluation rates without requiring the manually specified bounding box anchors used in competing methods.