Step-by-step
WeedAI accepts weed image datasets containing crops and weeds from all agricultural contexts. Your dataset should include images with annotations identifying the different weed and crop species.
Supported Annotation Formats
What is AgContext?
Writing down the conditions in which the data were collected and how it was collected are important for improving model development and allowing others to search. So we developed AgContext as an appended JSON to the MS COCO annotation format. It contains information on:
- Crop species and growth stage in which data were collected
- Background conditions including soil cover, soil colour, soil residue and location
- Camera type, lens, angle and speed
- Weather and other relevant collection information
Step-by-Step Upload Process
-
Prepare your dataset
Collect your images, review your data and annotate in the correct format. Remember to record the agcontext and dataset metadata.
-
(Recommended) Upload to Roboflow
Roboflow does a great job checking for image duplicates, missing annotations, small annotations, annotations outside image boundaries etc. These will all cause errors when uploading to the WeedAI platform - so we recommend you upload to Roboflow first so that your dataset has been checked. Alternatievly you can do this yourself.
-
Begin WeedAI Upload
Download data from Roboflow in Pascal VOC or COCO formats. Over on WeedAI Log in with your Google account and start the upload process.
-
Add annotations
Add your annotations and wait to press submit once they have loaded. Press next and map your annotations to species names. Use 'UNSPECIFIED' if you have a general 'weed' class. Scientific names must be in LOWERCASE. If you have growth stage or plant parts annotated, use the additional/optional field.
-
Create/Upload AgContext
Here is where you can create your AgContext file described above. If you already have one, simply use the button at the bottom to upload the stored .JSON file.
-
Create/Upload Metadata
Dataset metadata such as title, authors, funding, DOI are added here.
-
Upload images
For datasets over 100 images, we recommend using .zip files. Select Upload Zip File from the drop down list, and drag and drop your zip file onto the area.
-
Submit
Once the zip file has been uploaded, press submit and the upload process will begin. If successful, it will be reviewed by our team and you will receive a notification email once it has been indexed.