WasteDump-Dataset: YREB and Enhanced Global Dumpsite Annotations for Unregulated Waste Dump Segmentation

This dataset supports the paper "Label-Efficient Mapping of Unregulated Waste Dumps via Mixed Supervision with Feature Masked Recovery".

It contains two parts:

Part 1 - YREB Dataset: 16,429 samples (256x256 patches) from 11 administrative regions along the Yangtze River Economic Belt, China. Includes high-resolution satellite images (~0.26m spatial resolution), pixel-level binary masks (binary TIFF format, 0=background, 255=waste dump), and image-level labels. Pixel-level masks were manually annotated by expert interpreters using ENVI software.

Part 2 - Enhanced Global Dumpsite Annotations: Pixel-level segmentation masks (binary TIFF format, 0=background, 255=waste dump) for the Global Dumpsite dataset (Sun et al., 2023, Nature Communications), spanning 7 countries with 0.3-1.0m spatial resolution. The original images are available from the source paper. Our contribution is the fine-grained pixel-level delineation performed using Labelme software.

Code available at: https://github.com/rsshenli-geoai/SAM-FMR

------------------------------------------------------------Version Note (Major Revision Update)------------------------------------------------------------

This revised Supplementary Material corresponds to the major-revision version of the manuscript. Compared with the previously archived version, this update adds:  - Parameter sensitivity analysis of FMR (Table S5)  - Training-stage computational complexity analysis (Table S6)  - Representative failure-case visualizations on the YREB spatially     independent test set (Fig. S8)

The cost-performance analysis is updated to use the same-framework fully supervised reference. The evaluation pipeline is standardized with a unified micro-averaging implementation for F1-score, Precision, and Recall; this affects only the reporting of these auxiliary metrics in some tables. The IoU values, which serve as the primary metric, remain unchanged.

The Supplementary Material provides extended quantitative results, qualitative visualizations, sensitivity and efficiency analyses, and representative failure cases. See the document itself for the full section-by-section contents.