Settings and Their Meanings

Dandere2x primarily works via matching blocks between images, mostly through I-Frame and P-Frame compression.

Block Size

Block size refers to how big blocks should be in compression. In the image above, the block for the 'I-frame' is 3x1 Pacman-balls in size, while the furthest right 'I-frame' is 1x1 Pacman-balls in size.

For dandere2x, block sizes are dictated in pixels. The smaller the block size, the better the results, but exponentially gets slower the smaller it gets. The larger the block size, the worse the results, but is generally much faster.

Image:

Black box: 30x30 block in a 1080p image.
Red Box: 60x60 block in a 1080p image.

The way Dandere2x works, the smaller the block size, the slower the process takes. This stems from the fact that 'bled pixels' are added to compensate for the neural network creating a bleeding effect. The developer highly recommends the block size does not go below 15 and keeps it at 20 himself.

Image Quality

Since dandere2x works by 'recycling' blocks from previous frames, dandere2x uses 'image quality' as a metric of how acceptable a previous block has to be for it to be recycled. It does this by creating a compressed image based on a value between 1-100, similar to how photoshops quality slider works. If a block is matched with a previous frames block, and the match looks better than the 'compressed image-quality' block, the block is kept.

Image: Photoshop Quality Slider, which is very similar to how Dandere2x's image compression works.

Another way to articulate this is:

1. Find a matched block in a previous frame

2. How well does the matched block look when compared to the original block?

3. How well does the original block look when compared to the compressed block?

4. Does the comparison between the previous block and the original look better than the compressed block? If so, use it. If not, don't.