MoCaE: Mixture of Calibrated Experts Significantly Improves Object Detection
This addresses performance degradation in combining object detectors for researchers and practitioners, offering a method to enhance existing models without new architectures, though it is incremental as it builds on mixture-of-experts and calibration techniques.
The paper tackles the problem of degraded performance when naively combining expert object detectors by identifying miscalibration as the primary cause, and proposes a Mixture of Calibrated Experts approach that improves object detection on COCO by up to ~2.5 AP and achieves state-of-the-art results of 65.1 AP on COCO test-dev.
Combining the strengths of many existing predictors to obtain a Mixture of Experts which is superior to its individual components is an effective way to improve the performance without having to develop new architectures or train a model from scratch. However, surprisingly, we find that naïvely combining expert object detectors in a similar way to Deep Ensembles, can often lead to degraded performance. We identify that the primary cause of this issue is that the predictions of the experts do not match their performance, a term referred to as miscalibration. Consequently, the most confident detector dominates the final predictions, preventing the mixture from leveraging all the predictions from the experts appropriately. To address this, when constructing the Mixture of Experts, we propose to combine their predictions in a manner which reflects the individual performance of the experts; an objective we achieve by first calibrating the predictions before filtering and refining them. We term this approach the Mixture of Calibrated Experts and demonstrate its effectiveness through extensive experiments on 5 different detection tasks using a variety of detectors, showing that it: (i) improves object detectors on COCO and instance segmentation methods on LVIS by up to $\sim 2.5$ AP; (ii) reaches state-of-the-art on COCO test-dev with $65.1$ AP and on DOTA with $82.62$ $\mathrm{AP_{50}}$; (iii) outperforms single models consistently on recent detection tasks such as Open Vocabulary Object Detection.