DenseNet-PyTorch

Overview

This repository contains an op-for-op PyTorch reimplementation of Densely Connected Convolutional Networks.

Table of contents

• DenseNet-PyTorch
  • Overview
  • Table of contents
  • Download weights
  • Download datasets
  • How Test and Train
    • Test
    • Train model
    • Resume train model
  • Result
  • Contributing
  • Credit
    • Densely Connected Convolutional Networks

Download weights

• Google Driver
• Baidu Driver

Download datasets

Contains MNIST, CIFAR10&CIFAR100, TinyImageNet_200, MiniImageNet_1K, ImageNet_1K, Caltech101&Caltech256 and more etc.

• Google Driver
• Baidu Driver

Please refer to README.md in the data directory for the method of making a dataset.

How Test and Train

Both training and testing only need to modify the config.py file.

Test

• line 29: model_arch_name change to densenet121.
• line 31: model_mean_parameters change to [0.485, 0.456, 0.406].
• line 32: model_std_parameters change to [0.229, 0.224, 0.225].
• line 34: model_num_classes change to 1000.
• line 36: mode change to test.
• line 89: model_weights_path change to ./results/pretrained_models/DenseNet121-ImageNet_1K-30a6e303.pth.tar.

python3 test.py

Train model

• line 29: model_arch_name change to densenet121.
• line 31: model_mean_parameters change to [0.485, 0.456, 0.406].
• line 32: model_std_parameters change to [0.229, 0.224, 0.225].
• line 34: model_num_classes change to 1000.
• line 36: mode change to train.
• line 50: pretrained_model_weights_path change to ./results/pretrained_models/DenseNet121-ImageNet_1K-30a6e303.pth.tar.

python3 train.py

Resume train model

• line 29: model_arch_name change to densenet121.
• line 31: model_mean_parameters change to [0.485, 0.456, 0.406].
• line 32: model_std_parameters change to [0.229, 0.224, 0.225].
• line 34: model_num_classes change to 1000.
• line 36: mode change to train.
• line 53: resume change to ./samples/densenet121-ImageNet_1K/epoch_xxx.pth.tar.

python3 train.py

Result

Source of original paper results: https://arxiv.org/pdf/1608.06993v5.pdf)

In the following table, the top-x error value in () indicates the result of the project, and - indicates no test.

Model	Dataset	Top-1 error (val)	Top-5 error (val)
densenet121	ImageNet_1K	25.02%(25.57%)	7.71%(8.03%)
densenet161	ImageNet_1K	23.80%(22.92%)	6.85%(6.44%)
densenet169	ImageNet_1K	22.58%(24.42%)	6.34%(7.19%)
densenet201	ImageNet_1K	22.33%(23.12%)	6.15%(6.64%)

# Download `DenseNet121-ImageNet_1K-30a6e303.pth.tar` weights to `./results/pretrained_models`
# More detail see `README.md<Download weights>`
python3 ./inference.py 

Input:

Output:

Build `densenet121` model successfully.
Load `densenet121` model weights `/DenseNet-PyTorch/results/pretrained_models/DenseNet121-ImageNet_1K-30a6e303.pth.tar` successfully.
tench, Tinca tinca                                                          (99.53%)
barracouta, snoek                                                           (0.35%)
armadillo                                                                   (0.04%)
croquet ball                                                                (0.01%)
bolete                                                                      (0.01%)

Contributing

If you find a bug, create a GitHub issue, or even better, submit a pull request. Similarly, if you have questions, simply post them as GitHub issues.

I look forward to seeing what the community does with these models!

Credit

Densely Connected Convolutional Networks

Andrew Howard, Mark Sandler, Grace Chu, Liang-Chieh Chen, Bo Chen, Mingxing Tan, Weijun Wang, Yukun Zhu, Ruoming Pang, Vijay Vasudevan, Quoc V. Le, Hartwig Adam

Abstract

Recent work has shown that convolutional networks can be substantially deeper, more accurate, and efficient to train if they contain shorter connections between layers close to the input and those close to the output. In this paper, we embrace this observation and introduce the Dense Convolutional Network (DenseNet), which connects each layer to every other layer in a feed-forward fashion. Whereas traditional convolutional networks with L layers have L connections - one between each layer and its subsequent layer - our network has L(L+1)/2 direct connections. For each layer, the feature-maps of all preceding layers are used as inputs, and its own feature-maps are used as inputs into all subsequent layers. DenseNets have several compelling advantages: they alleviate the vanishing-gradient problem, strengthen feature propagation, encourage feature reuse, and substantially reduce the number of parameters. We evaluate our proposed architecture on four highly competitive object recognition benchmark tasks (CIFAR-10, CIFAR-100, SVHN, and ImageNet). DenseNets obtain significant improvements over the state-of-the-art on most of them, whilst requiring less computation to achieve high performance. Code and pre-trained models are available at this https URL .

[Paper]

@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},
  pages={4700--4708},
  year={2017}
}