RS-Neg Benchmark

We introduce RS-Neg, the first benchmark for evaluating negation understanding in remote sensing MLLMs across region-level to scene-level tasks. The benchmark includes VQA, MCQ, referring expression grounding, FloodNet, and scene classification annotations, covering both negated and non-negated evaluation settings.

Github Link: mumu011/NeFo.git

Image Sources

This repository contains annotation files only. The related images should be downloaded from the original dataset sources: