Task: We will be focusing on identifying different species of plants, so we need to choose the best dataset to do so. The larger the range of plant species, the better, but we can also focus on plants native to North America/Vancouver if we need to choose a smaller subset.

App idea is similar to this application that I found on Github:

https://github.com/UrbanPlanter/urbanplanterapp

The dataset I would go with is likely the Herbarium one (we would need to significant preprocessing that I could work on for next week - turning all the images into black-and-white, boundary detection, apply different convolutions and edge smoothing etc.)

Herbarium Dataset

• There is a dataset for 2021, 2022 competitions
• Flattened plant, doesn’t exactly represent what a plant would look like in real life
• Roughly 15,501 vascular plant species
• Would need to get rid of any species that didn’t have alot of images
• Needs alot of preprocessing (denoise, could be converted to a MNIST problem)

https://www.kaggle.com/competitions/herbarium-2022-fgvc9/data

https://www.kaggle.com/code/ihelon/herbarium-2021-exploratory-data-analysis/notebook

Flower Dataset

• 4000+ images
• 5 species: chamomile, tulip, rose, sunflower, dandelion
• Possibly why it won’t work: everyone knows what these flowers are already, there’s only 5 species in this dataset
• Lots of Kaggle project examples

https://www.kaggle.com/datasets/alxmamaev/flowers-recognition

Leaf Dataset

• 1,584 images of leaf specimens (16 samples each of 99 species)
• Possibly why it won’t work: not as many species as the Herbarium dataset, not alot of images per species

https://www.kaggle.com/competitions/leaf-classification/data