UESBTC/DenseNet-Tensorflow

Implementing by Tensorflow, train and evaluate DenseNet based on CIFAR-100 dataset.

DenseNet-Tensorflow

Not finished yet
Implementing by Tensorflow, it trains and evaluates DenseNet based on CIFAR-100 dataset. We have trained the network and release the weight files. The accuracy reaches 81.7% after training for 300 epoches.

How to Use

1. Download CIFAR-100 dataset(the Python version) from here
2. Unpack it to ./DenseNet-Tensorflow
3. Run python3 DenseNet.py

Reference

• Densely connected convolutional networks ZL Gao Huang, KQ Weinberger, L van der Maaten - ARXIV eprint, 2016
• Original implementation