NAIP data embedded with Clay v1.5 (rev2)

This repository has the embeddings using Clay vq.5 for the entire NAIP archive. See details on README.md License: Clay [CC-By](https://creativecommons.org/licenses/by/4.0/)

Product Details

Visibility: Public
Owner: Clay
Created: 22 Nov 2024
Last Updated: 3 Apr 2025

Product Contents

root

README

This repository contains all of NAIP on AWS data embedded with Clay v1.5.

Source: This data has been created by Development Seed under contract with Clay and comptute support by AWS.

Contact: For feedback and questions, please file a ticket on the model repo of Clay or email us at contact@madewithclay.org

Data Source: NAIP on AWS naip-analytic:

Includes NAIP imagery from 2010 through 2021.
The data in this bucket is Original Imagery in Geotiff format 4-band (RGB + NIR)

Model source: Embeddings generated from inference with Clay v1.5:

Embeddings have 1024 dimensions and correspond to the "class" embedding that is used alongside the patch embeddings at the end of the encoder.
Each tile is split into tiles of size 256x256, and the attention patch size inside Clay v1.5 is 8x8 px.
Inference run was done on November '24 on AWS using g4 and g6 EC2, for roughly 60k GPU hours, and costing around $65k.

Embeddings License: Clay CC-By

Format:

Folder structure follows the same as NAIP folder structure.
File format is parquet, with two columns: geometry and embeddings

Usage example

1import duckdb
2path = "https://data.source.coop/clay/clay-v1-5-naip-2/az/2015/100cm/rgbir_cog/32110/m_3211002_se_12_1_20150530.parquet"
3d = duckdb.read_parquet(path)
4df = d.to_df()
5df.head()
1import duckdb
2path = "https://data.source.coop/clay/clay-v1-5-naip-2/az/2015/100cm/rgbir_cog/32110/m_3211002_se_12_1_20150530.parquet"
3d = duckdb.read_parquet(path)
4df = d.to_df()
5df.head()