Am I missing something? As far i learn about EfficientNet that makes it diff from previous architecture is- Compound Scaling – means instead of scaling depth, width, or resolution individually scale all three using a fixed set of coefficients.
1
0
1
Replies
MBConv (Mobile Inverted Bottleneck Convolution) – It is the main building block of EfficientNet. It expands channels → applies depthwise conv → compresses them back.
1
0
0
SE (Squeeze and Excitation) Block – a lightweight attention mechanism that helps a CNN understand which channels are important and reweight them accordingly (obv this is only an overview). Swish Activation Func– used instead of ReLU. Regularization – DropConnect / Stochastic
0
0
0
