I should use a pretrained model, after that the model should be fed low res quality image the output of the model will be a high res image. The models output should be a high res image.
In one hot encoding, we could have a corner case, for example, let's suppose that we have a dataset of cars and it's one feature is "make" of the car. Now let's suppose that after one hot encoding "make" feature we get 20 columns for that feature. Now let's suppose we one hot encode the same feature for test set as well which gives us 19 columns instead of 20(some make model is missing in the test set). Now if we try to run some classifiers it won't work because the dimensions of test and training does not match, we can perform some preprocessing on the data if the test set is available but what if the test set is not available?