Closed magicse closed 1 year ago
I understand the issue at hand, and it appears that the pre-trained model is not suitable for production use. The training dataset consists of only 800 images, leading to unstable results when the model cannot identify certain portions of an image. By reducing the model's scope and utilizing a more extensive dataset, significant improvements can be made. I conducted a test using a different, larger dataset to restore VHS-like videos, as demonstrated in this link, and the results were more dependable.
I am confident that using a more extensive dataset can enhance the artifact removal model's performance, but unfortunately, the associated training time and cost are prohibitively high for a simple proof-of-concept project.
I try to restore old film 1978 And i get artifacts (small squares and some time green-yellow fragments) on frames. Please help me how to avoid this. It take so much time to calc and after that i see artifacts...