Skip to content

Reading and writing CIFAR file format. and, you can try TensorFlow's CIFAR-10 tutorials with the original dataset.

License

Notifications You must be signed in to change notification settings

TakeshiOkamoto/CIFAR-Converter

Repository files navigation

CIFAR Converter

Reading and writing CIFAR file format. and, you can try TensorFlow's CIFAR-10 tutorials with the original dataset.
(CIFARファイルの読み書き。そして、TensorFlowのCIFAR-10サンプルをオリジナルデータセットで試せます。)

You only preparing images.
(あなたは画像を用意するだけです。)

How to use (使い方)

1. Creation of training and evaluation file (訓練/評価ファイルの作成)

Create with "IMAGE to CIFAR" of cifar.exe.
(cifar.exeの「IMAGE to CIFAR」で作成します。)

CIFAR file format is no distinction between training and evaluation file.
(CIFARのファイルフォーマットでは訓練、評価ファイルの区別はありません。)

2. File placement (ファイルの配置)

Place the following files in the same folder.
(次のファイルを同じフォルダに配置します。)

File
cifar10.py
cifar10_eval.py
cifar10_ini.py
cifar10_input.py
cifar10_train.py
try.ipynb
Training file (Created in Chapter 1 / 1章で作成)
Evaluation file (Created in Chapter 1 / 1章で作成)

3. Edit settings (設定の編集)

Fit the cifar10_ini.py file to your environment.
(cifar10_ini.pyファイルを各自の環境にあわせます。)

4. Run training (訓練の実行)

python cifar10_train.py --max_steps=100000

5. Run evaluation (評価の実行)

python cifar10_eval.py

6. Try model (モデルを試す)

Just run "try.ipynb" with Jupyter Notebook.
(後は、Jupyter Notebookで「try.ipynb」を実行するだけです。)

but, please change the file name of test.jpg.
(但し、test.jpgのファイル名は変更してください。)

Mac / Linux

cifar.exe can be executed on Windows.
(cifar.exeはWindowsで実行可能です。)

Mac / Linux users have source code, please compile with "Lazarus".
(Mac / Linuxの方はソースコードがありますので、"Lazarus"でコンパイルして下さい。)

However, the operation is unconfirmed.
(但し、動作は未確認です。)

Contact

English Can understand, only 3-year-old level :-)

Other

MNIST Converter

About

Reading and writing CIFAR file format. and, you can try TensorFlow's CIFAR-10 tutorials with the original dataset.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published