Augmenting Hand-Drawn Art with Global Illumination Effects through Surface Inflation

Automatic Illumination of Flat-Colored Drawings by 3D Augmentation of 2D Silhouettes

“…Our illumination algorithm generates visually credible results (Fig. 1,4 and 5), similar in quality to far more ressouceintensive recent approaches based for instance on convolutional neural networks [4] (Fig. 5l,m).…”

“…Guiding lines (in magenta) can be seen on (e,j). For the sake of comparison, image (m) coming from [4] is shown, a recent method based on convolutional neural networks.…”

“…2. Nonetheless, our illumination results largely compete with more time-consuming methods based, for example, on convolutional neural networks [3,4,5]. 3.…”

“…[2]) and for the illumination of fixed or animated colorized drawings. For the latter, various formalisms can be found in the litterature: neural networks and GANs [3,4,5], interactive multiview 2.5D approaches [6,7], convolution on integral lines [8], treebased region representation [9], analysis of silhouettes partitions [10], of their skeletons [11] or of hatching lines added by the user [12]. In these state-of-the-art methods, user interaction is sometimes allowed to guide the algorithms towards a more personalized illumination result (e.g., to correct errors in the algorithmic estimation of the 3D scene geometry).…”

Polarized Object Surface Reconstruction Algorithm Based on RU-GAN Network

“…That is, assuming that the two gradients on the surface of the object are U and V , then the normal vector should be perpendicular to both gradients, i.e., the vectors are multiplied to 0. The gradients in the U and V directions can be solved by the height map; inspired by the literature [ 20 ], this paper solves the height map by constructing an energy function between the surface normal vector and the surface gradient, as follows. where the expressions for the gradients in the U -direction and V -direction are as follows.…”