A naukri.com initiative
Medium
3w
61
Image Credit: Medium
My First Deep Learning Project: Building a Naruto Image Classifier with FastAI
- As a beginner in deep learning and anime enthusiast, creating a Naruto image classifier served as an engaging starter project.
- The project involved collecting images using the duckduckgo_search Python library and utilizing Kaggle for model building and training.
- Git was used for version control, Visual Studio for developing a user-friendly interface, and Hugging Face Spaces for deploying the model.
- Data preparation included dataset cleaning, organization, and setting up data augmentation pipelines with FastAI tools.
- Transfer learning with a pre-trained CNN, specifically ResNet34, was used for model training.
- The model achieved an accuracy of approximately 89.8% with 123 correct predictions out of 137 validation images.
- Visualizing the top images with the highest loss revealed insights for improving the model.
- Utilizing FastAI’s ImageClassifierCleaner helped enhance dataset quality by identifying and removing problematic images.
- The deployment process involved exporting the model, creating a Gradio interface, and using Hugging Face Spaces for public accessibility.
- Visual Studio Code facilitated the deployment process, offering a smooth workflow from development to deployment.
- Challenges faced during the project included dataset quality, model architecture selection, and workflow optimization.
Read Full Article
3 Likes
For uninterrupted reading, download the app