Welcome to my Flower Image Classifier project! This venture was part of my Udacity capstone experience, showcasing my journey in training a neural network to classify images of flowers into 102 different categories.
In this script, I poured my efforts into training a new neural network. I opted for the VGG16 architecture, a powerful choice for image classification. The training spanned 10 epochs, with a batch size of 64 for training and 32 for both validation and testing.
Once the model was trained, I crafted a script to predict the classes of new images. This script loads the checkpoint saved during training and uses the model to predict the class of a given image.
This file holds the heart of my project—the saved state of the trained model. It encapsulates the model's architecture, weights, the mapping of classes to indices, and other crucial details.
Make sure to have the following dependencies installed:
- Python
- PyTorch
- NumPy
- Matplotlib
- PIL (Pillow)
Building this project wasn't all smooth sailing. Here are some challenges I encountered and how I tackled them:
GPU memory errors gave me a headache. I learned to tread carefully, using torch.cuda.empty_cache() and adjusting memory allocations when necessary.
Consistency in PyTorch versions across different platforms is key. I made sure to document version requirements to avoid unexpected hiccups.
Training deep neural networks takes time. I explored optimization strategies, delved into parallelization, and even considered cloud-based GPU resources for a quicker turnaround.
Implementing robust error handling became my North Star. Print statements and detailed logging were my allies in catching and resolving issues early in the development process.
After 10 epochs, my model achieved an impressive accuracy of 85% or more on the validation set. The journey was enriching, filled with experimentation and learning.
A huge shoutout to Udacity for providing the project framework and guidance. Also, immense gratitude to the PyTorch and open-source communities for contributing to the success of this project.
Here's to classifying flowers with the model I crafted! 🌸🌺🌼