2023 Digital AgAthon Labor Challenge Project

Based off of the PyTorch Torchvision object detection finetuning tutorial. Uses a pre-trained Mask R-CNN model that is finetuned to detect apples in orchards. The dataset used for traning "MinneApple: A Benchmark Dataset for Apple Detection and Segmentation"

• Torchvision Object Detection Finetuning Tutorial - https://pytorch.org/tutorials/intermediate/torchvision_tutorial.html#torchvision-object-detection-finetuning-tutorial
• MinneApple: A Benchmark Dataset for Apple Detection and Segmentation - https://conservancy.umn.edu/handle/11299/206575

Example

In the first image the masks were drawn over the test image using GIMP. The second image is the image used to test.

Last four images are examples of the masks detected on the test image and used to get the first image.

Guide

First, clone this repository,

• Install PyTorch and required dependencies, can be found in the torchvision object detection tutorial. Torchvision utilities are included in this repository.
• To test if your gpu is conneted

python gpu.py

• To train the model on your the appletrain dataset

python training.py

• To demo and test the model on an image from the dataset

python test.py

• load.py contains the model and dataset information.

To do

The detection isn't the most accurate as we were only given 48 hours to complete the project and train it. Due to hardware constraints we weren't able to train extensively.

• fix the output to draw masks over the images
• fix the number of masks outputted in demo