Receptive Field
The receptive field of a feature is the part of the original input that can affect it.
Early convolutional features see small local patches. Deeper features can depend on wider regions because they combine earlier features.
This is one reason convolutional networks can build hierarchy:
small local patterns -> larger patterns -> task-specific features
The receptive field is not the same as "what the model understands." It is the region that could influence a feature. Whether the model uses that region well is a training and architecture question.
Enter 1 if receptive field means the input region that can affect a feature, or 2 if it means the final loss value.
Compute it first, then check your number.
Enter 1 if deeper convolutional features can depend on wider input regions, or 2 if they must always see a smaller region.
Compute it first, then check your number.