Skip to content

Latest commit

 

History

History
34 lines (20 loc) · 957 Bytes

README.md

File metadata and controls

34 lines (20 loc) · 957 Bytes

CVSDemo

CIFAR-10 Demos in PyTorch

This repo contains demos used in the Computer Vision Systems course at BUTE (BMEVIIIMA07).

REQUIREMENTS:

  • Python 3
  • PyTorch 3.2 from www.pytorch.org
  • torchvision pip install torchvision
  • progressbar2 pip install progressbar
  • visdom pip install visdom

The first demo uses a simple neural network training implemented in plain PyTorch using nothing but autograd. You can run this demo by:

python net_plain.py

The second demo trains a neural network on random data using five different learning rate settings.

python net_torch.py

The third demo trains a small convolutional neural network on CIFAR-10. This achieves about 70% validation accuracy.

python convnet.py

The fourth demo trains a DenseNet169 net on CIFAR-10. This achieves ~95.6% validation accuracy.

python cifarTrain.py

python cifarTest.py

CREDITS:

DenseNet implementation by https://github.com/kuangliu/pytorch-cifar