EfficientDet: Scalable and Efficient Object Detection

The paper systematically studies neural network architecture design choices for object detection and introduces two key optimizations. The first is a weighted bi-directional feature pyramid network (BiFPN) that enables easy and fast multi-scale feature fusion. The second is a compound scaling method that uniformly scales resolution, depth, and width for the backbone, feature network, and box/class prediction networks at the same time. Combined with EfficientNet backbones, these yield a new detector family called EfficientDet.

Across a wide spectrum of resource constraints, EfficientDet consistently achieves much better efficiency than prior detectors. Using a single model at single scale, EfficientDet-D7 reaches state-of-the-art 52.2 AP on COCO test-dev with 52M parameters and 325B FLOPs, while being 4x-9x smaller and using 13x-42x fewer FLOPs than previous detectors. This mattered as model efficiency became increasingly important in computer vision, enabling strong detection under tight compute budgets.