Agriculture-Vision Database


The dataset used in this challenge is a subset of the Agriculture-Vision dataset [1]. The challenge dataset contains 21,061 aerial farmland images captured throughout 2019 across the US. Each image consists of four 512x512 color channels, which are RGB and Near Infra-red (NIR). Each image also has a boundary map and a mask. The boundary map indicates the region of the farmland, and the mask indicates valid pixels in the image. Regions outside of either the boundary map or the mask are not evaluated.

This dataset contains six types of annotations: Cloud shadow, Double plant, Planter skip, Standing Water, Waterway and Weed cluster. These types of field anomalies have great impacts on the potential yield of farmlands, therefore it is extremely important to accurately locate them. In the Agriculture-Vision dataset, these six patterns are stored separately as binary masks due to potential overlaps between patterns. Users are free to decide how to use these annotations.

Each field image has a file name in the format of (field id)_(x1)-(y1)-(x2)-(y2).(jpg/png). Each field id uniquely identifies the farmland that the image is cropped from, and (x1, y1, x2, y2) is a 4-tuple indicating the position in which the image is cropped. Please refer to our paper for more details regarding how we construct the dataset.


The challenge dataset contains images, boundaries and masks for train, val and test set. It also contains labels for the train and val set only. The dataset .tar.gz file is around 4.4 GB. Dataset terms can be found here. Please contact us to get access.


Agriculture-Vision: A Large Aerial Image Database for Agricultural Pattern Analysis

Mang Tik Chiu*, Xingqian Xu*, Yunchao Wei, Zilong Huang, Alexander Schwing, Robert Brunner, Hrant Khachatrian, Hovnatan Karapetyan, Ivan Dozier, Greg Rose, David Wilson, Adrian Tudor, Naira Hovakimyan, Thomas S. Huang, Honghui Shi

UIUC, IntelinAir, University of Oregon