Implemented a custom ResNet-34 variant for CIFAR-10 image classification with a focus on balancing classification accuracy and computational efficiency.
Overview
The project builds a lightweight deep learning model suitable for constrained environments. It keeps the residual learning structure of ResNet while adapting the implementation for smaller image inputs and practical experimentation.
Technical Focus
- Custom PyTorch implementation of residual blocks and ResNet-34 architecture
- CIFAR-10 training and evaluation workflow
- Model efficiency considerations for deployment-oriented experiments
- Notebook-based experimentation for reproducible model iteration
Repository
The full implementation is available on GitHub.