Invo-ResNet: Rethinking the Most Used Backbone in Computer Vision

ResNet has emerged as a widely adopted backbone in the field of Computer Vision. This research introduces a novel approach to augment conventional Convolution Neural Networks (CNNs) for image classification by incorporating involution layers, which exhibit inverse characteristics to convolution layers. By integrating these involution layers into the ResNet architecture, a remarkable reduction of nearly 200 times in model parameters is achieved. Despite this substantial parameter reduction, the model’s accuracy only experiences a marginal drop of approximately 5% when compared to the original ResNet, both trained on standard datasets. Moreover, a comparative analysis with the lightweight MobileNetV2 model demonstrates the involution-based approach’s superior parameter efficiency. This research presents an efficient alternative to traditional CNNs, contributing to advancements in memory utilization and optimization of training costs. All the Codes can be found in this link: https://github.com/de-bang/INVO_RESNET.git .