Cheaper Pre-training Lunch: An Efficient Paradigm for Object Detection

In this paper, we propose a general and efficient pre-training paradigm, Jigsaw pre-training, for object detection. Jigsaw pre-training needs only the target detection dataset while taking only 1/4 computational resources compared to the widely adopted ImageNet pre-training. To build such an efficient paradigm, we reduce the potential redundancy by carefully extracting useful samples from the original images, assembling samples in a Jigsaw manner as input, and using an ERF-adaptive dense classification strategy for model pre-training. These designs include not only a new input pattern to improve the spatial utilization but also a novel learning objective to expand the effective receptive field of the pre-trained model. The efficiency and superiority of Jigsaw pre-training are validated by extensive experiments on the MS-COCO dataset, where the results indicate that the models using Jigsaw pre-training are able to achieve on-par or even better detection performances compared with the ImageNet pre-trained counterparts.