ssd notes and code

13.7. Single-shot multi-frame detection (SSD) — hands-on deep learning 2.0.0 documentation

1. Then we need to classify ℎwq anchor boxes. If the fully connected layer is used as the output, it is easy to cause too many model parameters.

This sentence means that if the final output to a value hwq requires hwq parameters, for example, from x_1*w_1+x_2*w_2+..x_hwq*w_hwq, so hwq parameters are required -> convolutional layer

2.