Second
- Citation Author(s):
-
Kunping Yang
- Submitted by:
- Hao Cui
- Last updated:
- DOI:
- 10.21227/nsbb-ar76
- Data Format:
- Research Article Link:
25 views
- Categories:
- Keywords:
Abstract
SECOND is a well-annotated benchmark for SCD. It collects bitemporal aerial images covering Hangzhou, Chengdu, Shanghai and other cities. The number of images totals 4662 pairs, of which 2968 pairs are used for training and 1694 pairs are used for testing. Each image has a size of 512 × 512 pixels with a spatial resolution from 0.5m to 3m. SENCOND provides unchanged masks and land cover masks in changed areas. It focuses on 6 common land cover classes, i.e., non-vegetated ground surface, tree, low vegetation, water, buildings and playgrounds. These 6 classes form 30 frequently semantic change types. Meanwhile, SENCOND is imbalanced, and the changed samples account for 17.17% of the total samples.
Instructions:
SECOND is a well-annotated benchmark for SCD. It collects bitemporal aerial images covering Hangzhou, Chengdu, Shanghai and other cities. The number of images totals 4662 pairs, of which 2968 pairs are used for training and 1694 pairs are used for testing. Each image has a size of 512 × 512 pixels with a spatial resolution from 0.5m to 3m. SENCOND provides unchanged masks and land cover masks in changed areas. It focuses on 6 common land cover classes, i.e., non-vegetated ground surface, tree, low vegetation, water, buildings and playgrounds. These 6 classes form 30 frequently semantic change types. Meanwhile, SENCOND is imbalanced, and the changed samples account for 17.17% of the total samples.
