For years I've been trying to enhance the old "Shenmue Premiere" video that Sega published on their website more than a year before "Shenmue" was released. The video it's almost 2 hours long and it includes lots of interesting things, for example a ton of amazing footage of Beta and unreleased scenes of the first two games. Unfortunately, the original RealMedia format file from 1999 is 160x112 at 15fps (and frame rate is not even stable).
Image quality is so low and frame rate is so choppy that simply upscaling is not enough to make the video watchable. In recent years, artificial intelligence applied to video enhancement has increased its reliability, and recently I have decided to give it a try to enhance the "Shenmue Premiere" video. So here are the results.
Step 1: The original file. As I said, is 160x112 at 15fps:
Step 2: Using Artificial Intelligence frame interpolation, I've doubled the frame rate to 30 fps but keeping the original frame size of 160x112. The result is a more pleasant viewing experience but resolution is terribly low.
Step 3: Using the interpolated video from above and Artificial Intelligence upscaling, I've upscaled the frame size six times, getting a final result of 960x672 at 30fps. Of course, since the original file was so small and so compressed, the result is very irregular, and artificial intelligence upscaling is far from perfect. Some shots (especially closeups of faces) look very good but others look terrible. Still, I think it's now a much more pleasant viewing experience.
UPDATE 11/01/2021 - Step 4:
Using the same AI interpolation and AI upscaling processes, I've used the alternative video file provided by user MestreZiming as a source. The file had more problems of repeated frames (probably result of the original re-encoding) so there are more odd frames, specially in fast movement sequences. Nevertheless, the final result is very slightly better, with more natural colors and less compression artifacts:
Some sample frames:
Now we only need subtitles If someone wants to make subtitles I can add them to the video. Even though the video is long, there are long scenes with no dialog and some of them have already been translated from the "Project Berkley" GD, I think.
Image quality is so low and frame rate is so choppy that simply upscaling is not enough to make the video watchable. In recent years, artificial intelligence applied to video enhancement has increased its reliability, and recently I have decided to give it a try to enhance the "Shenmue Premiere" video. So here are the results.
Step 1: The original file. As I said, is 160x112 at 15fps:
Step 2: Using Artificial Intelligence frame interpolation, I've doubled the frame rate to 30 fps but keeping the original frame size of 160x112. The result is a more pleasant viewing experience but resolution is terribly low.
Step 3: Using the interpolated video from above and Artificial Intelligence upscaling, I've upscaled the frame size six times, getting a final result of 960x672 at 30fps. Of course, since the original file was so small and so compressed, the result is very irregular, and artificial intelligence upscaling is far from perfect. Some shots (especially closeups of faces) look very good but others look terrible. Still, I think it's now a much more pleasant viewing experience.
UPDATE 11/01/2021 - Step 4:
Using the same AI interpolation and AI upscaling processes, I've used the alternative video file provided by user MestreZiming as a source. The file had more problems of repeated frames (probably result of the original re-encoding) so there are more odd frames, specially in fast movement sequences. Nevertheless, the final result is very slightly better, with more natural colors and less compression artifacts:
Some sample frames:
Now we only need subtitles If someone wants to make subtitles I can add them to the video. Even though the video is long, there are long scenes with no dialog and some of them have already been translated from the "Project Berkley" GD, I think.
Last edited: