Here collect projects related to anime / anime face. The main focus is fun rather than novel contribution.
tdrussell/IllustrationGAN based on DCGAN (ICLR 2016)
Towards the Automatic Anime Characters Creation with Generative Adversarial Networks
- collect dataset from Getchu
- leverage proper, empirical application of DRAGAN (Deep Regret Analytic Generative Adversarial Networks On Convergence and Stability of GANs)
- using praticable GAN training strategies online service: MakeGirlsMoe with lot of setting
CartoonGAN: Generative Adversarial Networks for Photo Cartoonization
In this paper, we propose a solution to transforming photos of real-world scenes into cartoon style images
(1) cartoon styles have unique characteristics with high level simplification and abstraction, and (2) cartoon images tend to have clear edges, smooth color shading and relatively simple textures, which exhibit significant challenges for texture-descriptor-based loss functions Two novel losses suitable for cartoonization are proposed:
(1) a semantic content loss, which is formulated as a sparse regularization in the high-level feature maps of the VGG network to cope with substantial style variation between photos and cartoons, and
(2) an edge-promoting adversarial loss for preserving clear edges.
We further introduce an initialization phase, to improve the convergence of the network to the target manifold.
styleGAN Generate with noise + latent space control in different scale.
This WaiFu does not exist
Making Anime Faces With StyleGAN - Gwern.net
StyleGANv2
Quality of image usually is very high. sometime there is broken/weird output.
U-GAT-IT: Unsupervised Generative Attentional Networks with Adaptive Layer-Instance Normalization for Image-to-Image Translation| OpenReview
tensorflow | pyTorch | reddit
- new attention module that could handle geometric (shape) changes, with an auxiliary classifier
$\mu_s$provide attention map for decoder - learnable normalization: AdaLIN (Adaptive Layer-Instance Normalization) with parameters
$\gamma, \beta$computed from attention map
There are 2 implemetation/app with U-GAT-IT models.
selfie2anime site
找了4張照片玩玩看
1: 臉形blur
1,2: 眼睛不太對稱，動漫的大眼睛比較難維持結構吧。中二病異色瞳XD
3: 側面失敗了，沒了一只眼
4: 在這照片中,我髮尾染了紫藍色。我本身想着動漫多彩色頭髮，不過好像辨認不出藍色部份(也許被我的黑色衣服影響了)，直接斷開了。線條很分明。可惜沒有其他我染髮期間的照片。(這張是在>20人的家庭聚會截出來......)
整體上髮型保持得很好。我算是比較幸運了，demo的sample比我還要奇怪
雖然圖2出了藍紫異色瞳，但整體雙眼的顏色是相近的。
P.S. 大概因為dataset多女, 女生效果比男生好。讓男友試一下，徹底悲劇了。
blog: Iterating on an idea
Selfie2WaiFu
Based on U-GAT-IT official pretrain model.
Selfie2WaiFu 自動檢測人臉, 圖3把背景的竹當成人臉，玩不了。
個人感覺WaiFu線條較實(分明, 顏色接近黑色)，眼睛沒有崩比較對稱，但眼睛以外的物件(眼鏡框, 頭髮)Selfie2Anime完整些。
也許只是個例。同一張畫裏畫風、線條風格相當統一，但輸入的照片會用哪種畫風是隨機。古早畫風線條較實，近年畫風比較保留原圖和容易模糊。可能可以用一些conditional cGAN控制畫風?(用年代/制作組區分)
Waifu Labs
Quality of image is high
repeatedly pick 1 image from 16 grid
I guess it is like StyleGAN? After picking 1 image, randomly generate 16 noise for the feature at next scale and append to picked image.
First Order Motion Model for Image Animation (NIPS 2019)
Project
With reference source image and driving frame, generate animation with keypoints detection
not related to anime generation, but I saw some control anime face with real video on twitter