Semantic Segmentation of Remote Sensing Images With Sparse Annotations

Training Convolutional Neural Networks (CNNs) for very high resolution images requires a large quantity of high-quality pixel-level annotations, which is extremely labor- and time-consuming to produce. Moreover, professional photo interpreters might have be involved guaranteeing the correctness annotations. To alleviate such burden, we propose framework semantic segmentation aerial based on incomplete where annotators are asked label few pixels with easy-to-draw scribbles. exploit these sparse scribbled FEature Spatial relaTional regulArization (FESTA) method complement supervised task an unsupervised learning signal that accounts neighbourhood structures both in spatial feature terms.