Universal Style Transfer via Feature Transforms

• Universal style transfer must reconcile generalization, visual quality, and efficiency when applying style characteristics to content images.Source: §1 · p. 1
• Multi-level stylization matches style statistics across feature levels, while the approach also demonstrates high-quality visual results and universal texture synthesis.Source: Figure 1 · p. 2, §1 · p. 2
• The method transforms intermediate content features so their statistical characteristics match those of style features at corresponding layers.Source: §1 · pp. 1–2
• Whitening and coloring transforms directly match content feature statistics to those of a style image in deep feature space.Source: §1 · p. 2
• A fixed VGG encoder and pre-trained decoder networks reconstruct stylized images through simple feed-forward operations after feature transformation.Source: §1 · p. 2, Figure 1 · p. 2

• Analogy-based methods often require content-style image pairs depicting the same scene type, limiting their scalability to arbitrary style images.Source: §2 · pp. 2–3
• Optimization-based arbitrary stylization matches deep-feature correlations but is inefficient because it requires iterative optimization.Source: §2 · p. 3
• Feed-forward transformation networks improve efficiency but commonly require training one network per style because of limited generalization.Source: §2 · p. 3
• Later multi-style methods use style-specific units, filters, or normalization parameters, while local feature swapping targets arbitrary style transfer.Source: §2 · p. 3
• The proposed approach combines feed-forward efficiency with generalization and visual quality on arbitrary styles, using WCT rather than adaptive instance normalization.Source: §2 · p. 3

Reconstruction decoder

• The reconstruction framework uses a fixed VGG-19 encoder and trained symmetric decoders to invert features back to RGB images.Five decoders are trained for features from Relu_X_1, X=1,2,3,4,5.Source: §3.1 · p. 3

Whitening and coloring transforms

• WCT whitens content features and colors them with style statistics so their covariance matches the style feature covariance.Whitening removes feature correlations after centering; coloring applies the style covariance and re-centers with the style mean.Source: §3.2 · p. 4
• Whitened features retain global content structures while removing style-related information, including detailed stroke patterns.The features are inverted with a reconstruction decoder to visualize what content and style information remains.Source: §3.2 · p. 4, Figure 2 · p. 4

Feature transform comparison

• WCT captures salient style patterns more faithfully than channel-wise histogram matching, which transfers global color but breaks patterns and misrepresents local structures.The difference is attributed to WCT modeling correlations between feature channels through the covariance matrix.Source: §3.2 · pp. 4–5

User control

• The stylization strength is user-controllable by blending the transformed style feature with the original content feature using the style weight α.The blended feature is then passed to the decoder.Source: §3.2 · p. 5

Multi-level stylization

• The multi-level pipeline processes higher VGG layers first for coarse stylization, then lower layers to refine details and low-level information.Higher layers capture complicated local structures or salient patterns, whereas lower layers carry colors and further improve details.Source: Figure 3 · p. 4, §3.3 · p. 5, Figure 5 · p. 5

• The evaluation compares stylization methods across paintings, abstract styles, and styles with obvious texton elements, using α = 0.6 for our method.Source: Figure 6 · p. 6
• Our method captures salient style patterns while preserving and stylizing key content components such as bridges, eyes, and mouths.Source: §4.2 · p. 7
• Lower covariance matrix difference Ls indicates stylized results closer to the given style statistics, and our method achieves lower Ls.Source: §4.2 · p. 7
• A user study with 80 subjects and 1,200 votes found that our method received the most votes for better stylized results.Source: §4.2 · p. 7

Efficiency

• The approach is efficient but slightly slower than [27] [15] because its eigenvalue decomposition is implemented on the CPU.Source: §4.2 · pp. 7–8

User Controls

• The method provides scale, weight, and spatial controls through feed-forward operations, including mask-based stylization of specified regions.Source: §4.2 · p. 8

Texture synthesis

• For texture synthesis, repeated multi-level processing improves visual appeal, while sampled noise produces diverse results without the repetitive layouts observed for [27].Source: §4.3 · pp. 8–9
• Feature-space interpolation generates new mixed textural effects, whereas produces results described as overlaid combinations of two textures.Source: §4.3 · p. 9, Figure 10 · p. 9

• The algorithm avoids learning for each individual style by matching intermediate content and style feature statistics and correlations through whitening and coloring transforms.Source: §5 · p. 9
• A multi-level stylization pipeline incorporates information from all style levels, and the approach is also effective for texture synthesis.Source: §5 · p. 9