Photographic style transfer using gradient histogram matching

Please click on the thumbnails to see the full-resolution images and use the back button to return to this page.

Input image (HDR)

Results

• We demonstrate the results obtained with our approach once convergence has been reached (see paper for details).
• We provide the results of two other method for style transfer, produced by the authors themselves. The harmonization technique was not designed to handle HDR inputs. Minor artifacts may appear with such inputs but these would most likely be removed by fine tuning the settings. More importantly, the harmonization technique produces outputs that do not match the model style as well as our method, in particular, the local contrast is often too weak.
• We also provide the result obtained with the method of Bae et al. [2006]. The output style is sometimes weak. This can be addressed by not applying the final step based on the Poisson equation. However, this comes at the cost of artifacts such as halos.

Model	Harmonization [Sunkavalli et al. 2010]	Two scale tone management [Bae et al. 2006]	[Bae et al. 2006] (no Poisson equation)	our method: gradient statistics