Tinyvision:-Building Ultra-Lightweight Models for Image Tasks(Part-1)

TinyVision explores how small image classifiers can get while remaining useful, replacing learned CNN filters with a handcrafted 10-channel feature pipeline (edges, textures, gradients, etc.). Part 1 covers a cat vs dog classifier using this fixed preprocessing approach and four compact architectures (Model0–Model3) with grouped convolutions, plus a partial CIFAR-10 analysis using purely compact CNN design without the filter bank.