Whether scaled version of boundary of training data or region proposal fed to the pre-trained CNN model?

In the slide as well as during the lecture about Feature extraction in RCNN model it is mentioned that the proposed regions are cropped to fit the input dimensions of any CNN model. But isn’t this an error in the lecture slide as we feed in only the scaled version of the boundary of training data (as mentioned in the lecture RCNN: Training) to the pre-trained CNN model?