https://github.com/YixuanLi/densenet-tensorflow

DenseNet Implementation in Tensorflow
https://github.com/YixuanLi/densenet-tensorflow

densenet tensorflow

Last synced: about 1 month ago
JSON representation

DenseNet Implementation in Tensorflow

• Host: GitHub
• URL: https://github.com/YixuanLi/densenet-tensorflow
• Owner: YixuanLi
• License: gpl-3.0
• Created: 2016-09-22T14:52:49.000Z (about 8 years ago)
• Default Branch: master
• Last Pushed: 2019-05-07T17:47:07.000Z (over 5 years ago)
• Last Synced: 2024-08-01T22:49:54.733Z (4 months ago)
• Topics: densenet, tensorflow
• Language: Python
• Size: 248 KB
• Stars: 574
• Watchers: 23
• Forks: 197
• Open Issues: 10
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• awesome-very-deep-learning - Tensorflow Implementation
• project-awesome - YixuanLi/densenet-tensorflow - DenseNet Implementation in Tensorflow (Python)
• awesome-image-classification - unofficial-pytorch : https://github.com/YixuanLi/densenet-tensorflow
• awesome-image-classification - unofficial-pytorch : https://github.com/YixuanLi/densenet-tensorflow

README

        
# DenseNet-tensorflow

This repository contains the tensorflow implementation for the paper [Densely Connected Convolutional Networks](http://arxiv.org/abs/1608.06993).

The code is developed based on Yuxin Wu's implementation of ResNet (https://github.com/ppwwyyxx/tensorpack/tree/master/examples/ResNet).

Citation:

     @inproceedings{huang2017densely,

          title={Densely connected convolutional networks},

          author={Huang, Gao and Liu, Zhuang and van der Maaten, Laurens and Weinberger, Kilian Q },

          booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},

          year={2017}

      }

## Dependencies:

+ Python 2 or 3

+ TensorFlow >= 1.0

+ [Tensorpack] (https://github.com/ppwwyyxx/tensorpack)

+ OpenCv-Python

## Train a DenseNet (L=40, k=12) on CIFAR-10+ using

```

python cifar10-densenet.py

```

In our experiment environment (cudnn v5.1, CUDA 7.5, one TITAN X GPU), the code runs with speed 5iters/s when batch size is set to be 64. The hyperparameters are identical to the original [torch implementation] (https://github.com/liuzhuang13/DenseNet).

## Training curves on CIFAR-10+ (~5.77% after 300 epochs)

![cifar10](cifar10.png)

## Training curves on CIFAR-100+ (~26.36% after 300 epochs)

![cifar100](cifar100.png)

## Differences compared to the original [torch implementation] (https://github.com/liuzhuang13/DenseNet)

+ Preprocessing is not channel-wise, instead we use mean and variances of images.

+ There is no momentum and weight decay applied on the batch normalization parameters (gamma and beta), whereas torch vertison uses both momentum and weight decay on those.

## Questions?

Please drop [me](http://www.cs.cornell.edu/~yli) a line if you have any questions!