In this project, I develop a deep learning model and train it on a large set of flower image data in order to accurately predict the names of flowers in completely new images. You can find the full code here and below.
I had quite a bit of fun analyzing this beautiful British flowers data set that contains 102 categories of flowers with 40 to 258 images per class, you can read more about it here. Using Pytorch (which is an optimized tensor library for deep learning using GPUs and CPUs), I built a deep learning model that utilizes transfer learning to build upon and improve a pre-designed neural network and achieve results with 80% accuracy.
This model is designed to provide accurate results as well as the probabilities of the top 5 predictions. Once you feed a new picture into the model, it automatically shows you the image with the flower’s name as the title, and produces a bar chart that shows the top 5 predicted classes. Here are some examples:
In the Notebook below, I go through the process of obtaining the data pre-processing the images, building, training and testing the classifier, and lastly, drawing inference from it.
Overall, this was an excellent project that shows the power of image classification and the tremendous potential it has. One potential example is actually a project that I’d like to explore, which entails analyzing debris data in the aftermath of a natural or man-made disaster to help rescue efforts reach people faster. I will be searching for data sets which preferably have heat imaging, and if anyone has any insight, please feel free to reach out.
Here is a link to the code in full page mode: