— Many approaches for image segmentation rely on a 1 first low-level segmentation step, where an image is partitioned 2 into homogeneous regions with enforced regularity and adherence 3 to object boundaries. Methods to generate these superpixels have 4 gained substantial interest in the last few years, but only a few 5 have made it into applications in practice, in particular because 6 the requirements on the processing time are essential but are not 7 met by most of them. Here, we propose waterpixels as a general 8 strategy for generating superpixels which relies on the marker 9 controlled watershed transformation. We introduce a spatially 10 regularized gradient to achieve a tunable tradeoff between the 11 superpixel regularity and the adherence to object boundaries. 12 The complexity of the resulting methods is linear with respect 13 to the number of image pixels. We quantitatively evaluate our 14 approach on the Berkeley segmentation database and compare 15 it against the state-of-the-art.