A Comparative Study of Fourier Transform and CycleGAN as Domain Adaptation Techniques for Weed Segmentation - Code and Data
Description
This dataset contains the code and data to reproduce the experiments of the paper "A Comparative Study of Fourier Transform and CycleGAN as Domain Adaptation Techniques for Weed Segmentation". The data comes from the ROSE Challenge, a benchmarking competition of agricultural robots focused on autonomous weed destruction. The ROSE Challenge has been organized by the National Laboratory of Metrology and Testing (LNE) and the National Research Institute for Agriculture, Food, and the Environment (INRAE). Four teams participated in the ROSE field campaigns with different robots and camera systems. The teams' names are BIPBIP, PEAD, ROSEAU, and WeedElec. The images used in our experiments were collected in the years 2019 and 2021 in an experimental field at the INRAE research center located in Montoldre, France. The teams scanned maize (Zea mays) and bean (Phaseolus vulgaris) plants with four kinds of weeds (Lolium perenne, Sinapis arvensis, Chenopodium album, Matricaria chamomilla) under natural daylight conditions. The dataset is composed of RGB images (with different resolutions) and semantic segmentation masks to distinguish the crop, weed, and background classes. There are 1000 labeled images in total (125 per team and crop type) per year.