A Large-Scale Dataset and a New Method for RemoteSensing Traffic Object Segmentation
The researchers created NWPU-Traffic, a large-scale remote sensing dataset featuring instance-level annotations for four vehicle categories (cars, airplanes, ships, and trains) across 49 cities in 7 countries. They established a benchmark using popular segmentation networks and introduced a new segmentation method that utilizes spatial-channel preserving feature interaction alongside an adaptive feature decoder to handle varying object scales.
Why it matters — It provides a highly diverse, multi-country dataset and a specialized model that improves the accuracy of segmenting transport objects across complex, real-world environments, overcoming the category and scene limitations of previous datasets.