smallssd 0.0.5

Open source agricultural data

smallSSD: The small robot company's semi-supervised detection dataset

smallSSD is an open source agricultural semi-supervised object detection dataset, containing 960 images labelled with wheat and weed bounding boxes and 100,032 unlabelled images.

All images were collected by the Small Robot Company's Tom robot in 8 experimental fields with varying drill rates and fertilizer and herbicide application, and are available on Zenodo.

This repository returns a dataset, modelled off the torchvision datasets:

from torch.utils.data import DataLoader
from smallssd.data import LabelledData, UnlabelledData

labelled_loader = DataLoader(LabelledData())

By default, this code expects the labelled data to be in the data folder (and will automatically download it from Zenodo if it is not available there).

More in-depth examples on how to get started with this data are available in the benchmarks, where we train torchvision models against the data using both fully-supervised and pseudo labelling approaches.

Installation

smallSSD can be installed with the following command:

pip install smallssd

License

smallSSD has a Creative Commons Attribution-NonCommercial 4.0 International license.