2022 ACRE Cascade Campaign

In the 2022 ACRE 1st Cascade Campaign, participants are asked to segment RGB images to distinguish between crop, weeds, and background.

Data has been collected in real crop fields during 2019 and 2021 agricultural robotics competitions. The dataset comprises images captured by two agricultural robots in different moments and with different RGB cameras. Images consist of two kinds of crops (maize and bean) and multiple species of weeds. Participants are provided with labeled images to train their models, and they are asked to submit their hypothesis (segmented images) of a test dataset.

 

The competition is structured into three stages:

  1. Development: in this stage, participants are asked to develop a model to perform semantic segmentation of RGB images by training their models on the 2019 dataset.
  2. Generalization: in this stage, participants are asked to submit predictions of the unlabelled 2021 dataset by using their models trained on the 2019 dataset. Different environmental conditions and sensors’ settings require the models to have generalization capability. The generalization capability can be reached by applying for style transfer and/or domain adaptation techniques.
  3. Final: in this stage, participants are required to submit predictions of a new unlabelled 2021 test set. This stage is thought to submit the final model without major changes; thus, the duration is limited to three days and the number of submissions to three.

 

Competition timeline:

Stage 1 - Development opens: 23rd February 2022

Stage 2 - Generalization opens: 6th April 2022

Stage 3 - Final opens: 12nd May 2022

Stage 3 - Final closes: 15th May 2022

Algorithm submission deadline: 17th May 2022

Final rank: 23-27th May 2022

 

Register here

 

Competition timeline:

Stage 1: Development opens: February 23, 2022

Stage 2: Generalization opens: April 6, 2022

Stage 3: Final: May 12 -15, 2022

Algorithm submission deadline: May 17, 2022

Final rank: May 23-27, 2022

More information: 

acre(a)metricsproject.eu