Cloud detection from remotely observed data is a critical pre-processing step\nfor various remote sensing applications. In particular, this problem becomes\neven harder for RGB color images, since there is no distinct spectral pattern\nfor clouds, which is directly separable from the Earth surface. In this paper,\nwe adapt a deep pyramid network (DPN) to tackle this problem. For this purpose,\nthe network is enhanced with a pre-trained parameter model at the encoder\nlayer. Moreover, the method is able to obtain accurate pixel-level segmentation\nand classification results from a set of noisy labeled RGB color images. In\norder to demonstrate the superiority of the method, we collect and label data\nwith the corresponding cloud/non-cloudy masks acquired from low-orbit Gokturk-2\nand RASAT satellites. The experimental results validates that the proposed\nmethod outperforms several baselines even for hard cases (e.g. snowy mountains)\nthat are perceptually difficult to distinguish by human eyes.\n