To test the robustness of our method, the pictures in this stack were captured with a predefined delay (two 33ms frames were dropped between the capture of the source and reference images). This is a case where our sparse-to-dense registration fails (flip '1'-'3'). Apart from a few areas, such as the neon lights in the back, a single homography computed from SIFT matches works well in this case ('4'). However, our modification of Exposure Fusion is able to compensate for both SIFT's problems and ours ('5' and '6').