---

Waifu2x Comparisons

---

Waifu2x Projects:

1. nagadomi's waifu2x

• Supports Linux.

2. waifu2x-caffe

• Supports Windows.

3. WL-Amigo's waifu2x-converter

• Supports Windows.

Comparison Results:

• nagadomi's waifu2x seems to produce the cleanest results with the least amount of noise/artifacts.

• The upconv_7 art models perform better than the vgg_7 art models, and as such the upconv_7 art model was used for this example.

• The custom model was trained using nagadomi's waifu2x. The model was trained on images created by Neural-Style, and many of the associated style images.

• This model was trained from scratch with the art variant of the upconv_7 model.

• Both of the waifu2x-caffe results seem to produce noisier images with more artifacts.

• WL-Amigo's waifu2x-converter seems to produce to cleaner images than waifu2x-caffe, but it's still not as good as nagadomi's waifu2x in terms of artifacts/noise.

Notes:

• Each project's RGB model was used for each test. If an "Art" variant of the RGB model existed, then it was used.

• Keep in mind that there can be situations where you want more noise/artifacts, or less noise/artifacts.

• The original/starting image size before Waifu2x was used, is 1664x1637. The original image was UpScaled 2 times, to a resolution of 3328x3274.

• The Waifu2x UpScaled images then had a 512x256 piece cropped from their centers, which was then resized to 1000x512 and the annotations were added. Imagemagick was used for this process.

---