728x90
Challenge 1: efficient multi-scale feature fusion
Challenge 2: model scaling
3.2. Cross-Scale Connections
3.3. Weighted Feature Fusion
Unbounded fusion: However, since the scalar weight is unbounded, it could potentially cause training instability.
Softmax-based fusion: the extra softmax leads to significant slowdown on GPU hardware.
Fast normalized fusion: this fast fusion approach has very similar learning behavior and accuracy as the softmax-based fusion, but runs up to 30% faster on GPUs
4.1. EfficientDet Architecture
4.2. Compound Scaling
Backbone network – we reuse the same width/depth scaling coefficients of EfficientNet-B0 to B6 [39] such that we can easily reuse their ImageNet-pretrained checkpoints
BiFPN network
Box/class prediction network
Input image resolution
