//=time() ?>
Some experimentation with #SpaceQuest 6 graphics and #ESRGAN. It sometimes requires a little manual aliasing on some solid lines, but the result is quite good!
taking another crack at an ESRGAN model designed for the DKC series
this model is trained on vintage 3D renders, with color banding applied to the LR tiles. unsure which degree of banding produces the best results, or whether to train a 4x model and downscale the SR photos
the 2nd place winner of the NTIRE2020's 16x challenge, ciplab, used a method called "run ESRGAN twice"
training a 'sequel' to manga109attempt for ESRGAN.
JPG/dither friendly, trained on photos and traditional/digital art, 2x upscale (only nn/bicubic filtering should go beyond 2x imo), looks a bit soft but 'safe', doesn't radically alter the image
18 hours, 0 epochs ._.
Hmm think I still prefer ESRGAN (1) to SRFBN (2). ESRGAN adds a lot more grain, but that works better for my images. SRFBN also has some weird grid artifacts.
remastered the milk bar from majora's mask 3D using ESRGAN, materialize and blender. with help from @redtheundead. check it out here: https://t.co/G1EdL1txx7 #vrchat
Quest For Glory 4 added to my #ESRGAN upscale collection: https://t.co/B2JC8VREfB
#AINeuralNet #QuestForGlory #Sierra
So AI Neural Net upscaling is going to revolutionize remastering and recovering old games/graphics/photos/drawings. This is incredible. It's hard to believe it's a filter! This was done with ESRGAN using a Mang109 dataset-trained model. Days of hq4x and SuperEagle are numbered!