training FRCNN- requires_grad

[javadvahid]

in the book it says we freeeze the parameters but the requires_grad is set to True?
is that correct?
why?
shouldnt it be set to false like in RCNN?

class FRCNN(nn.Module):
def init(self):
super().init()
rawnet = torchvision.models.vgg16_bn(pretrained=True)
for param in rawnet.features.parameters():
param.requires_grad = True