UFOD: An AutoML framework for the construction, comparison, and combination of object detection models

摘要

Object detection models based on deep learning techniques have been successfully applied in several contexts; however, non-expert users might find challenging the use of these techniques due to several reasons, including the necessity of trying different algorithms implemented in heterogeneous libraries, the configuration of hyperparameters, the lack of support of many state-of-the-art algorithms for training them on custom datasets, or the variety of metrics employed to evaluate detection algorithms. These challenges have been tackled by the development of UFOD, an automated machine learning framework that trains several object detection algorithms (using different underlying frameworks and libraries), compares them, and finally selects the best model or ensembles them. Currently, the most well-known object detection algorithms have been included in our system, and new methods can be easily incorporated thanks to a high-level API. UFOD is available at https://github.com/ManuGar/UFOD/ ? 2021 Elsevier B.V. All rights reserved.