SJTUzhanglj / tensorpack

Neural Network Toolbox on TensorFlow

tensorpack

Neural Network Toolbox on TensorFlow

See some examples to learn about the framework:

Vision:

• DoReFa-Net: train binary / low-bitwidth CNN on ImageNet
• Train ResNet on ImageNet/Cifar10/SVHN
• InceptionV3 on ImageNet
• Fully-convolutional Network for Holistically-Nested Edge Detection(HED)
• Spatial Transformer Networks on MNIST addition

Reinforcement Learning:

• Deep Q-Network(DQN) variants on Atari games
• Asynchronous Advantage Actor-Critic(A3C) with demos on OpenAI Gym

Unsupervised Learning:

• Several Generative Adversarial Network(GAN) variants, including DCGAN, Image2Image, InfoGAN

Speech / NLP:

• LSTM-CTC for speech recognition
• char-rnn for fun

The examples are not only for demonstration of the framework -- you can train them and reproduce the results in papers.

Features:

Describe your training task with three components:

1. Model, or graph. models/ has some scoped abstraction of common models, but you can simply use any symbolic functions available in tensorflow, or most functions in slim/tflearn/tensorlayer. LinearWrap and argscope simplify large models (vgg example).

2. DataFlow. tensorpack allows and encourages complex data processing.

   • All data producer has an unified generator interface, allowing them to be composed to perform complex preprocessing.
   • Use Python to easily handle any data format, yet still keep good performance thanks to multiprocess prefetch & TF Queue prefetch. For example, InceptionV3 can run in the same speed as the official code which reads data by TF operators.

3. Callbacks, including everything you want to do apart from the training iterations, such as:

   • Change hyperparameters during training
   • Print some variables of interest
   • Run inference on a test dataset
   • Run some operations once a while
   • Send loss to your phone

With the above components defined, tensorpack trainer will run the training iterations for you. Multi-GPU training is off-the-shelf by simply switching the trainer. You can also define your own trainer for non-standard training (e.g. GAN).

The components are designed to be independent. You can use Model or DataFlow in other projects as well.

Dependencies:

• Python 2 or 3
• TensorFlow >= 0.11
• Python bindings for OpenCV
• other requirements:

pip install --user -r requirements.txt
pip install --user -r opt-requirements.txt (some optional dependencies, you can install later if needed)

• Enable import tensorpack (or use greadlink from coreutils brew package if you're on OSX):

export PYTHONPATH=$PYTHONPATH:`readlink -f path/to/tensorpack`

• Use tcmalloc if running with large data