geographicgoogle/RSRCCremote-sensinggeospatialchange-detectionvqamultimodalbenchmarksatelliteimage-textgoogle-researchsemantic-change

RSRCC Remote Sensing Change Comprehension Benchmark

Free

Open dataset

Sample structure: 100 / 100

3 download links issued

Seller: DataBazaar

Already have an account? Log in

Agent? Connect your account →

Source, license and coverage

Supplier documentation. These claims are separate from the automated sample score. A listing edit date is not a data freshness date.

License: Apache-2.0
Source / creator: google/RSRCC
Collection method: The dataset was constructed via a retrieval-augmented Best-of-N ranking pipeline: multi-temporal remote sensing image pairs depicting regional changes are paired with candidate natural language descriptions/questions, and the highest-quality candidates are selected through retrieval-augmented ranking against reference evidence. The result is a benchmark targeting semantic (not just pixel-level) change comprehension, where models must reason about what changed and why rather than only localizing pixel differences.
Coverage start: Not documented
Coverage end: Not documented
Data last updated: Not documented
Update schedule: Not documented

Source documentation ↗

License terms ↗

Embedded binary payloads are explicitly replaced with byte-length descriptors; the preview preserves accompanying text and metadata. Language coverage is English only. The source paper documents the construction pipeline but buyers should validate geographic coverage, sensor diversity, temporal gap distribution, and class balance empirically before using for training or evaluation. As a Best-of-N retrieval-ranked dataset, answer phrasing reflects the ranker's preferences and may carry stylistic bias. License terms and any usage restrictions should be re-verified on the HF dataset page prior to commercial deployment.

Sample structure score: 100 / 100

This automated check describes the inspected sample, not factual accuracy, legal rights, representativeness, or the quality of the entire dataset. It is not a customer rating.

Assessed 10 sample records (JSON) on 2026-10-09. All records in the provided sample were checked.

Check	Points	Evidence
Populated cells	50 / 50	30 of 30 top-level cells contain a value. Null, absent and blank values count as missing; zero and false count as populated.
Consistent value types	30 / 30	30 of 30 populated cells match their column's most common observed type. Types are inferred, not checked against real-world truth.
Consistent record shape	20 / 20	10 of 10 records have the expected fields. CSV/TSV use the header width; JSON uses the union of observed keys.

Field-level findings and improvements

Check missing cells and mixed types below. Document intentional missing values or mixed types in your field descriptions. Do not fill legitimate unknowns with invented values just to increase this score.

Field	Missing cells	Most common type	Other populated types
before	0 / 10	object	0 / 10
after	0 / 10	object	0 / 10
text	0 / 10	string	0 / 10

How the score is calculated, its limitations, and how to correct an assessment →

About this data

Multimodal benchmark for semantic change understanding using multi-temporal satellite image pairs paired with natural language questions and answers. Supports visual question answering, change captioning, and multiple-choice tasks.

Retrieve with your agent or Python

Create an account and configure DATABAZAAR_API_KEY. This example retrieves free or already purchased data; it never makes a purchase. For a multi-file dataset, choose a file index from the manifest.

Download the Python example

python3 retrieve-dataset.py 65fb4289-df76-4c63-9d1a-5176bf52a4c9 --output dataset.bin

Full supplier documentation

## Overview RSRCC (Remote Sensing Regional Change Comprehension) is a multimodal benchmark from Google Research's RSFM (Remote Sensing Foundation Models) team for semantic change understanding in remote sensing. It pairs multi-temporal satellite/aerial image evidence with natural language questions and answers, constructed via retrieval-augmented Best-of-N ranking. The dataset sits in the 100K–1M sample range and is stored as an imagefolder with paired text annotations, supporting visual question answering, image-text-to-text generation, and multiple-choice evaluation. ## Schema - image_before — image — earlier-timestamp remote sensing image of the region - image_after — image — later-timestamp remote sensing image of the same region - question — string — natural language question about the regional change - answer — string — reference natural language answer - choices — list[string] — multiple-choice options (where applicable) - task_type — string — VQA / change captioning / multiple choice - (additional metadata fields per the HF imagefolder layout) ## Sources - Hugging Face: https://huggingface.co/datasets/google/RSRCC — published by Google Research - Associated paper: arXiv:2604.20623 (RSRCC) - License: as declared on the HF dataset page (Google Research release) ## Methodology The dataset was constructed via a retrieval-augmented Best-of-N ranking pipeline: multi-temporal remote sensing image pairs depicting regional changes are paired with candidate natural language descriptions/questions, and the highest-quality candidates are selected through retrieval-augmented ranking against reference evidence. The result is a benchmark targeting semantic (not just pixel-level) change comprehension, where models must reason about what changed and why rather than only localizing pixel differences. ## Known gaps & limitations Language coverage is English only. The source paper documents the construction pipeline but buyers should validate geographic coverage, sensor diversity, temporal gap distribution, and class balance empirically before using for training or evaluation. As a Best-of-N retrieval-ranked dataset, answer phrasing reflects the ranker's preferences and may carry stylistic bias. License terms and any usage restrictions should be re-verified on the HF dataset page prior to commercial deployment. ## Intended use & out-of-scope - IS for: evaluating and fine-tuning multimodal/VLM systems on remote-sensing change understanding, change captioning, and geospatial VQA; benchmarking RSFM-class models. - NOT for: pixel-accurate change-detection segmentation training (this is a semantic/language benchmark, not a mask dataset); leakage-sensitive benchmark training without deduplication against your eval splits. _Federated dataset: 49 parquet shards, 22.09 GB total. Queries and downloads stream through the DataBazaar API._ Original supplier listing: RSRCC: Remote Sensing Regional Change Comprehension Benchmark Google Research multimodal benchmark for semantic change understanding in remote sensing — multi-temporal satellite image pairs with natural language Q&A for VQA, change captioning, and multiple-choice tasks.

Schema

Name	Type	Description
before	STRUCT(bytes BLOB, path VARCHAR)	Earlier-timestamp remote sensing image as JPEG bytes and file path reference
after	STRUCT(bytes BLOB, path VARCHAR)	Later-timestamp remote sensing image of same region as JPEG bytes and file path reference
text	VARCHAR	Natural language question, reference answer, or task metadata concatenated as single string

Sample Data

Preview a sample of the data before downloading.

Public sample only. Sign in to retrieve the full dataset, including free datasets.

For AI Agents

Via MCP Server

# 1. Add to your agent's MCP config (claude_desktop_config.json or similar):
{
  "mcpServers": {
    "databazaar": { "command": "npx", "args": ["databazaar-mcp"] }
  }
}

# 2. Your agent can then call:
search_datasets({ query: "RSRCC Remote Sensing Change Co" })
// Found: 65fb4289-df76-4c63-9d1a-5176bf52a4c9
get_download_url({ dataset_id: "65fb4289-df76-4c63-9d1a-5176bf52a4c9" })  // free — sign in with MCP OAuth first

Via REST API

# Free dataset — sign in or use your account API key:
curl https://api.databazaar.io/datasets/65fb4289-df76-4c63-9d1a-5176bf52a4c9/download-url -H "Authorization: Bearer $DATABAZAAR_API_KEY"