# Light Stable Representations (LSR): Drone Orthomosaic Tiles at Three Times of Day with Canopy Height, MPG Ranch, Montana

## Facts

- Provider: MPG Ranch (Kyle Doherty, Erik Samose, Max Gurinas, Brandon Trabucco, Ruslan Salakhutdinov)
- License: CC-BY-4.0 (https://huggingface.co/datasets/mpg-ranch/drone-lsr)
- Size: 24 files
- Format: Parquet (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/mpg-ranch/drone-lsr
- Access: public, readable without an account
- Category: Earth Observation & Geospatial
- Source: https://huggingface.co/datasets/mpg-ranch/drone-lsr
- Page: https://scigantic.com/archive/a1962337-c1b9-544c-a914-a0a6d02050d0
- Keywords: Earth Observation & Geospatial, remote sensing, Parquet, Hugging Face

## About

Aerial RGB orthomosaic tiles of the same ground captured at three times of day (10:00, 12:00 and 15:00) on 7 November 2024 over Lower Partridge Alley at MPG Ranch, Montana, with a co-registered canopy height model. 620 tile sets of 1024 x 1024 pixels at 1.2 cm ground resolution; canopy height is an int32 grid in centimeters. Three configurations share tile ids and an 80/20 train/test split: default (the three images plus canopy height), dinov2_base (DINOv2 ViT-B/14 CLS and patch tokens per image) and dinov3_sat (DINOv3 ViT-L/16 satellite-pretrained, SAT-493M, CLS and patch tokens). Tiles with transient objects such as vehicles were removed. Built to train and test vision encoders whose features stay stable under changing illumination and shadows, and to compare natural-image and satellite-pretrained encoders for drone remote sensing of vegetation. Parquet, about 6.7 GB; load with datasets.load_dataset('mpg-ranch/drone-lsr', config). Licensed CC BY 4.0; cite Doherty, Samose, Gurinas, Trabucco and Salakhutdinov (MPG Ranch, 2024).

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("mpg-ranch/drone-lsr")
print(ds)
```

## Data structure

File types in sampled files: Parquet

Example file: `data/test-00000-of-00003.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/a1962337-c1b9-544c-a914-a0a6d02050d0
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/a1962337-c1b9-544c-a914-a0a6d02050d0
- Over MCP: `get_schema_card` with id `a1962337-c1b9-544c-a914-a0a6d02050d0` at https://api.scigantic.com/mcp

## Related topics

- [Earth Observation & Geospatial datasets](https://scigantic.com/topics/earth-observation-and-geospatial)
- [Parquet datasets](https://scigantic.com/topics/parquet)
- [Remote sensing datasets](https://scigantic.com/topics/remote-sensing)
