Sign In and Upload
We welcome new contributions of datasets of images with weeds already annotated.
Annotation Formats
Annotations can currently be uploaded in the following formats, for which we note the supported annotation layers:
Format | Classification | Bounding Boxes | Segmentation |
---|---|---|---|
WeedCOCO | ✔️ | ✔️ | ✔️ |
MS COCO | ✔️ | ✔️ | ✔️ |
Pascal VOC XML | ✔️ | ||
Color-coded segmentation masks | ✔️ |
We natively support WeedCOCO format which extends on MS COCO to specify a weed ID-oriented category naming scheme, to include agricultural context and schema.org/Dataset-compatible metadata. Uploads in other formats will be converted to WeedCOCO format in our uploader, which provides forms to enter agricultural context and metadata, as well as to map category names to our standardised nomenclature.
Licensing
Images and annotations are owned and copyright held by their respective contributors as defined in the metadata file. We require that contributors license their images and annotations under the CC-BY 4.0 licence (Creative Commons Attribution Required).
Uploaders must have the rights to the content that they upload, and must agree to release their content (images and annotations) under the terms of that licence.
In brief (and not a substitute for the License), the CC BY 4.0 License enables users to freely share and adapt the material for any purpose, even commercially, given appropriate attribution and that there are no additional restrictions made.
Upload Process
The upload tool will walk you through the upload process, which varies depending on the annotation type and format. The process consists of the following steps:
- Select the data annotation format.
- Upload the annotation file(s). Please check that your images and annotations match.
- If not WeedCOCO, confirm a mapping from the entered category labels to role (weed or crop) and species/taxon.
- If not WeedCOCO, upload the AgContext file or generate a new one by completing the online form.
- Enter publication-level metadata about the dataset and how to attribute it.
- Finally, upload the relevant images for the dataset
A Weed-AI administrator will review your submission. If the dataset is accepted you will be notified and can continue uploading new datasets. The new dataset will appear on the datasets page for other users to peruse.
Only one dataset per uploader can be in submission and review at any time.
Prepare for your upload
Some tips to prepare your dataset:
- Construct and save an AgContext (download it as JSON).
- Get a DOI and construct and save metadata.
- OPTIONAL: Get the annotations into MS COCO format. You might want to use the
weedcoco.importers
package of theweedcoco
Python library, or using a third party tool like Roboflow.
Data Quality Requirements
Images submitted should be unedited and of high quality. Submitted datasets that contain any of the following will be rejected by dataset reviewers:
- Irrelevant images (such as non-weed or crop imagery)
- Poor image quality (over/under exposed, blurry images)
- Augmented images (those that have been processed in some manner)
- Synthetic images (computer-based generation of image data)
- Unlabelled or poorly labelled data
- Images with personally identifiable content (such as faces and vehicle details)
- Images with explicit content
Currently only images that are new to the repository will be accepted.