And again about deinterlace

And again about deinterlace.

Here is a part from Death Note anime.
It is full interlace (you can see very small lines on edges of background) = every field is a frame.
Thats why DDV correctly deinterlace background.
But for japanese text:

  1. Its different from frame to frame.
  2. Some frames from the start there is a black line appeared near 4th japanese char.
    Same as I’v shown for Armitage anime.

Robust model does not produce such artifacts.

Also such artifacts do not appear if enlargement is used.

@Artur_Alekseev Thank you for sharing!

I think it is related to model training. Its not trained on source, that contain static content (like in this example is japanese text). Same thing for anime (that has exact images from frame to frame).
Try to deinterlace any progressive anime source as interlaced and you will get same black lines artifacts, that ofcourse should not appear.

@Artur_Alekseev Can you send the video as well? I believe that I only have the image sequences at this time.

Sent in private message.

1 Like

Thank you @Artur_Alekseev

We have confirmed that this is a model issue and is something that will take time to resolve. More than likely we will need to create a new model to handle cases like this.