LAE-1M (Locate Anything on Earth - 1 Million) is a large-scale open-vocabulary remote sensing object detection dataset introduced in the paper "Locate Anything on Earth: Advancing Open-Vocabulary Object Detection for Remote Sensing Community" (AAAI 2025).

It contains over 1M images with coarse-grained (LAE-COD) and fine-grained (LAE-FOD) annotations, unified in COCO format, enabling zero-shot and few-shot detection in remote sensing.

Dataset Details

Dataset Sources

Repository: GitHub - LAE-DINO
Paper: ArXiv 2408.09110, AAAI 2025
Project Page: LAE Website
Dataset Download: HuggingFace

Dataset Structure

Subset	# Images	# Classes	Format	Description
LAE-COD	400k+	20+	COCO	Coarse-grained detection (AID, EMS, SLM)
LAE-FOD	600k+	50+	COCO	Fine-grained detection (DIOR, DOTAv2, FAIR1M)
LAE-80C	20k (val)	80	COCO	Benchmark with 80 semantically distinct classes

All annotations are in COCO JSON format with bounding boxes and categories.

Uses

Direct Use

Open-Vocabulary Object Detection in Remote Sensing
Benchmarking zero-shot and few-shot detection models
Pretraining large vision-language models

Out-of-Scope Use

Any tasks requiring personal or sensitive information
Real-time inference on satellite streams without further optimization

Quick Start

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("jaychempan/LAE-1M", split="train")

# Access one example
example = dataset[0]
print(example.keys())  # image, annotations, category_id, etc.

# Show the image (requires Pillow)
from PIL import Image
import io

img = Image.open(io.BytesIO(example["image"]))
img.show()