Input normalisation issue

Is it possible to un-normalized inputs to petrained VGG-16.because if i pass unnormalized images the lower level features are not detected properly.But it is also recommended in neural style transfer to clamp data between 0,255 for better image quality.Is it possible to pass the image without normalization and achieve exication of lower level features.

Help !!!

This could be possible, if you retrain the model (which might be trickier using unnormalized inputs).
Note that the pretrained models were trained using normalized inputs. Unnormalized inputs are “new” to the model and thus the extracted features might be way off.

Sorry i didnt retain model part