Rethinking Transformer-based Set Prediction for Object Detection
This work addresses the slow training convergence and accuracy limitations of Transformer-based object detection models, which is a significant problem for researchers and practitioners using DETR-like architectures.
This paper investigates the slow convergence of DETR, a Transformer-based object detection method, identifying issues with its Hungarian loss and Transformer cross-attention. The authors propose TSP-FCOS and TSP-RCNN, which achieve faster convergence and significantly higher detection accuracy than DETR and other baselines.
DETR is a recently proposed Transformer-based method which views object detection as a set prediction problem and achieves state-of-the-art performance but demands extra-long training time to converge. In this paper, we investigate the causes of the optimization difficulty in the training of DETR. Our examinations reveal several factors contributing to the slow convergence of DETR, primarily the issues with the Hungarian loss and the Transformer cross-attention mechanism. To overcome these issues we propose two solutions, namely, TSP-FCOS (Transformer-based Set Prediction with FCOS) and TSP-RCNN (Transformer-based Set Prediction with RCNN). Experimental results show that the proposed methods not only converge much faster than the original DETR, but also significantly outperform DETR and other baselines in terms of detection accuracy.