Inspiration
As a plant biology major, identifying flowers is a skill that I find challenging and interesting at the same time. I was curious how accurately I could build a model to correctly classify flowers given a small number of labels.
What it does
The model takes an image of a flower (or any image) and labels it with one of five possible flower types.
How we built it
I used the Sequential class Keras to train a CNN with a dataset of about 3600 flower images and 5 possible labels.
Challenges we ran into
It was difficult to increase the the accuracy of the model no matter how many epochs were trained. Clearly, there was a different issue altogether, possibly the structure of the CNN or the quality of the training set.
Accomplishments that we're proud of
Not any in particular.
What we learned
Flower classification is a difficult task.
What's next for Flower Classification with CNN
- Improve quality of training data
- Increase number of labels (include more flowers)
- Try different number of layers and different number of nodes for layers.
- Try different optimizers.
- Try different libraries and algorithms such as YOLOv5 in PyTorch
Log in or sign up for Devpost to join the conversation.