Non-Local-Nets

This is an implement of Non-local neural networks for tensorflow version. Here, you can see the paper provided by Xiaolong Wang et.al.
Non-Local-Block shows below:

Requirements

python 2.7
tensorflow > 1.0.0
numpy
tqdm
cPickle(python2.7), pickle(python3.*)

Usages

Download Repo

$ git clone https://github.com/nnuyi/Non-Local-Nets
$ cd Non-Local-Nets

Datasets

In this repo, I mainly focus on MNIST, CIFAR10 datasets.

MNIST: You are not required to download MNIST datasets since I use tensorflow mnist tool to obtain this datasets, so you just run this repo like the following steps.
CIFAR10: You are required to download CIFAR10 datasets here, unzip it and store it in './data/cifar10/', note that CIFAR-10 python version is required. You can unzip it in './data/cifar10/' using the following command:
```
$ tar -zxvf cifar-10-python.tar.gz
# you will see that data_batch_* are stored in './data/cifar10/cifar-10-batches-py/'
```

TODO:

In the following time, I will test it in cifar100 etc datasets to see wheather this networks can work well

Training

MNIST

If this is first time you run the repo, it will download MNIST automatically it will cost about 5 to 10 seconds, please wait for a moment. After that, you need not to download MNIST again since it have been downloaded at first time. Just see the following instructions for training phase:

  # MNIST is the default option
  $ python main.py --is_training=True --is_testing=False
  
  # If GPU options is avaiable, you can use it as the instruction shows below:
  $ CUDA_VISIBLE_DEVICES=[no] python main.py --is_training=True --is_testing=False
  
  # notes: [no] is the device number of GPU, you can set it according to you machine
  $ CUDA_VISIBLE_DEVICES=0 python main.py --is_training=True --is_testing=False

CIFAR10

  $ python main.py --is_training=True --is_testing=False --datasets=cifar10 --input_height=32 --input_width=32 --input_channels=3
  
  # If GPU options is avaiable, you can use it as the instruction shows below:
  $ CUDA_VISIBLE_DEVICES=[no] python main.py --is_training=True --is_testing=False --datasets=cifar10 --input_height=32 --input_width=32 --input_channels=3
  
  # notes: [no] is the device number of GPU, you can set it according to you machine
  $ CUDA_VISIBLE_DEVICES=0 python main.py --is_training=True --is_testing=False --datasets=cifar10 --input_height=32 --input_width=32 --input_channels=3

Testing

MNIST

In this repo you can will see the testing phase during training phase since I ran the test_model codes to test its performance per 5 epochs. If you have finished training phase and want to test it, just see the following instructions:

  # MNIST is the default option
  $ python main.py --is_training=False --is_testing=True

CIFAR10

  $ python main.py --is_training=False --is_testing=True --datasets=cifar10 --input_height=32 --input_width=32 --input_channels=3

Results

MNIST

After about 1 epochs or less, you can see that the testing accuracy rate can reach to more than 96.00%. And training accuracy rate can reach to 98.36%.
After about 100 epochs or less, you can see that the testing accuracy rate can reach to more than 99.39%. And training accuracy rate can reach to 99.91%. I run this repo in Geforce GTX 1070 GPU, it cost 8 seconds per epoch.

CIFAR10

After about 100 epochs or less, you can see that the testing accuracy rate can reach to more than 81.22%. And training accuracy rate can reach to 99.16%. I run this repo in Geforce GTX 1070 GPU, it cost 30 seconds per epoch.