DeepCanola: Phenotyping Brassica Pods Using Semi-Synthetic Data and Active Learning

Dataset accomapnying the publication: DeepCanola: Phenotyping Brassica Pods Using Semi-Synthetic Data and Active Learning by Van Vliet, Atkins et al. We provide model weights, training and vlidation datasets, as well as phenotype outputs. Each file/folder is outlined below: deepcanola.pth - Model weights for the final Model 4, named DeepCanola generated_datasets - Datasets generated at each stage of the active learning process, datasets 1-4 Each folder is an iteration of the active learning process, inside each folder are the generated images and associated annotations stored in the COCO format. real_world_datasets - Real-world datasets used for either creation of the pod pools or validation. Datasets include: br9 - Ordered and disordered dataset of images with generated pod length data of the ordered images stored in the `br9_gt_lengths.csv` file br11 - Ordered and disordered dataset of images only br17 - Ordered dataset with ground-truth of images with length annotations collected in ImageJ and stored in the `BR017 POD SCAN DATA.csv` file. misc - Dataset of miscellaneous images including Brassica napus from Rothamstead and brassica relatives data_generation_pools - Pools used to generate semi-synthetic data at each step of the active learning process. Pools include: background_pool - Created background images to be selected at random by the semi-synthetic data generation script pod_pools - Pools of pods used in the semi-synthetic data generation process. Pod pools include: br9 - 673 pods with associated masks br9 and br17 - 673 + 332 pods with associated masks deepcanola_outputs - Phenotype data outputs generated by DeepCanola. Each output is stored as a .csv file of both length measurements of each pod (with _objects.csv suffix), and average length measurements per image (with _averages.csv suffix). Outputs include: br9_ordered br9_disordered br17