Versatile CNN (for image classification, can be optimized to provide efficient TinyML solutions : good validation accuracy yet low complexity)
Code is avalailable to run in Google COLAB (or in Kaggle platform)
V-CNN is a less constrained architecture composed of an arbitrary number of macro-layers, each can be assigned any degree of non-linearity. The idea of (non)linear macro-layer is simillar to NL-CNN but V-CNN allows arbitrary # of filters per each macro-layer and arbitrary nonlinearity. It also allows to add any desired number of hidden dense layers in the output classifier. NL-CNN and XNL-CNN models are now special cases of the V-CNN.
News: (Apr. 25, 2024): A Pytorch implementation was added: https://github.com/radu-dogaru/V-CNN/blob/main/vcnn_pytorch.py
News: (June 28, 2024): A newer Tiny-ML model namned VRES-CNN was derived (it includes V-CNN as a special case). More detalis here: https://github.com/radu-dogaru/vres-cnn