This project proposes a modified ResNet architecture for image classification tasks with the aim of reducing the number of trainable parameters while maintaining acceptable accuracy. The goal is to develop a more efficient image classifier suitable for deployment on resource-constrained devices, such as Internet of Things (IoT) devices. The proposed model achieved a test accuracy of 93.4% on the CIFAR-10 dataset with just 2.79 million trainable parameters, demonstrating the effectiveness of the approach. The results of this project contribute to ongoing research on efficient deep learning models and have potential implications for real-world applications where model efficiency is critical.
The file modifiedResNet_best.ipynb is the final submission for Deep learning mini project by vb2279, ap7982 and smk8939.