This repository provides a PyTorch implementation of Bayesian convolutional neural networks where intractable posterior probability distributions over weights are inferred using variational inference. The approach replaces point estimates with learned distributions, enabling uncertainty quantification in deep learning models. It demonstrates performance equivalent to frequentist inference on standard image classification benchmarks including MNIST, CIFAR-10, and CIFAR-100. The implementation includes both standard Bayes by Backprop and flipout variants for computational efficiency.