input image size in trainFaste​rRCNNObjec​tDetector

I try to use trainFasterRCNNObjectDetection in Matlab. As i understanding, in original faster R-CNN paper, input size of CNN first layer is image size, for example, 256*256. But in Matlab example;

https://kr.mathworks.com/help/vision/examples/object-detection-using-faster-r-cnn-deep-learning.html?searchHighlight=fast%20r%20cnn%20object&s_tid=doc_srchtitle they recommend to use small object size such as 32*32. I don't understand this part. For applying CNN, input layer has full image size? right? Pleas help me!