dsanno / chainer-cifar

Chainer implementation of CIFAR-10 dataset training

Implementation of CIFAR-10 dataset classification using Chainer

Requirements

• Python 2.7
• Chainer 2.0.0
• Cupy 1.0.0
• Pillow
• matplotlib

How to use

Download dataset

$ python src/download.py

Convert dataset

$ python src/dataset.py

This command makes some dataset files:

• Image dataset files
  • dataset/image.pkl: Only mean value is subtracted
  • dataset/image_zca.pkl: Applied ZCA whitening
  • dataset/image_norm_aca.pkl: Applied Contrast normalization and ZCA whitening
• Label file
  • dataset/label.pkl

Train

Example:

$ python src/train.py -g 0 -m vgg -b 128 -p vgg --optimizer adam --iter 300 --lr_decay_iter 100

While training the following files are saved.

• model parameter file <prefix\>.model
• loss and error log file <prefix\>_log.csv
• loss curve image <prefix>_loss.png
• error curve image <prefix>_error.png

Options:

• -g (--gpu) <int>: Optional
  GPU device ID. Negative value indicates CPU (default: -1)
• -m (--model) <model name>: Optional
  Model name that is one of the following (default: vgg)
  • cnn: Simple CNN
  • cnnbn: Simple CNN with Batch Normalization
  • vgg: VGG like model
  • residual: Residual Network
• -b (--batch_size) <int>: Optional
  Mini batch size (default: 100)
• -d (--dataset) <file path>: Optional
  File path of image dataset file (default: dataset/image.pkl)
• -l (--label) <file path>: Optional
  File path of label file (default: dataset/label.pkl)
• -p (--prefix) <str>: Optional
  Prefix of saved model file. (default: <model name>_<optimizer name>)
• --iter <int>: Optional
  Training iteration (default: 300)
• --save_iter <int>: Optional
  Iteration interval to save model parameter file. 0 indicates model paramete is note saved at fixed intervals. Note that the best accuracy model is always saved even if this parameter is 0. (default: 0)
• --optimizer <str>: Optional
  Optimizer name (sgd or adam, default: sgd)
• --lr <float>: Optional
  Initial learning rate for SGD (default: 0.01)
• --alpha <float>: Optional
  Initial alpha for Adam (default: 0.001)
• --lr_decay_iter <int>: Optional
  Iteration interval to decay learning rate. Learning rate is decay to 1/10 at this intervals. (default: 100)
• --weight_decay <float>: Optional
  Weight decay (default: 0.0001)
• --res_depth <int>: Optional
  Depth of Residual Network. Total number of residual blodks is res_depth * 3 and total number of layers is res_depth * 6 + 2 (default: 18)
• --skip_depth: Optional
  Use stochastic depth in Residual Network if set.
• --seed <int>: Optional
  Random seed (default: 1)

License

MIT License