Supplemental Material: Automatic Content-aware Projection for 360◦ Videos

Figure 1 to Figure 7 show the additional results of our contents-aware projection method with the results of other projection methods to compare the performance qualitatively. For clear comparison of the proposed contents-aware projection method, we collected additional 360◦ images from the web. We used automatically extracted line segments and salient points as inputs for Carroll’s method and our projection algorithms for the fair comparison. Overall, rectilinear projection preserves every line in images but causes stretching distortions at the borders. In stereographic projection, conformality of objects is preserved, but all the lines except radial lines are bent. Pannini projection [3] preserves vertical lines, but the horizontal lines are not well preserved in many cases. Carroll’s method [1] preserves only the extracted line segments. Our optimized Pannini projection preserves horizontal lines better, but a little stretch at the borders. Our model-interpolated optimized Pannini method prevents these stretches occurred at the border of the projected images while preserving the conformality of the salient regions well — it preserves lines and conformality of salient regions simultaneously (thanks to the parameter optimization) over the whole image (thanks to model interpolation). In addition, although our method is optimization-based, our parameter optimization is very fast and takes about 1ms on a single CPU. In the figures, red arrows indicate noticeable distortions and green arrows indicate improvements by the proposed method.