Get a flying start

Data collection is expensive and this dataset is for you to lower your start cost for innovative projects in weed control. The dataset has a permissive license to enable rapid progress in research as well commercial projects.

Details on the dataset

While this is the official webpage of the OpenWeeds dataset it contains mainly overview information as well as being the repository for the data and annotations. For a more detailed description we refer to the base publication for the OpenWeeds dataset.

Tinker away

In addition to the annotated data and raw data we supply all the test code we have developed during the project. This includes code to communicate with CVAT, train and evalute models as well as for generating result graphs. The code is published on GitHub.

Overview of available resources

Approximately 250 000 annotations, details in the publications.

Annotation for detection

The clearly largest part of the annotation set consists of bounding box annotations of weeds together with their class labels / names.

Sparse segmentation annotation

Some data was annotated with sparse segmentation masks. A challenge which might require e.g. non-standard loss layers. This data was annotated by the WASP WARA Media & Language.

Camera calibration data

The project did not use this data but we supply checker board data captured by the project cameras for your calibration needs. Keep in mind that most data was captured from stereo pairs of cameras.

Code

Download of the dataset is done through code which gets the data from this website. There is also code for generation of result graphs etc.

Are you ready to continue your agricultural adventure?

Insert Coin