The dataset contains about 4000 images resized to 512x512. There are ~5000 fruit in total. Some images contain more than one fruit and thus more than one bounding box. The images are annotated using bounding boxes defined in a COCO format and each bounding box is tagged to one of three classes: fruit_healthy, fruit_brownspot and fruit_woodiness.

The images were collected from the Eastern, Western and Central regions of Uganda. The image capturing process involved guidance from National Crop Resources Research Institute (NaCRRI) passion fruit disease experts, who identified the disease manifestation in the fruits of passion fruit plants.

The objective of this challenge is to classify the disease status of a plant given an image of a passion fruit. You need to classify each fruit individually and not assume that all the fruit in the same image have the same status.

Files available for download:

• Train_Images.zip - contains the ~3000 resized (512x512) images.
• Test_Images.zip - contains ~ 1000 images resized (512x512) images.
• Train.csv - contains the image IDs that are in the train set and the label (fruit_healthy, fruit_brownspot or fruit_woodiness) target. You will use this dataset to select the images from Train_Images.zip to train your model on.
• Test.csv- contains image IDs for the test set. You will use this dataset to select the images from Test_Images.zip to apply your model to.
• SampleSubmission.csv - shows the submission format for this competition, with the ‘Image_ID’ column mirroring that of Test.csv. The order of the rows does not matter, but the names of the‘Image_ID’must be correct. You may submit a maximum of 4 bounding boxes per image.