Tensorpack Documentation

Tensorpack is a training interface based on TensorFlow, with a focus on speed + flexibility. TensorFlow is powerful, but has its own drawbacks: Its low-level APIs are too hard and complicated for many users, and its existing high-level APIs sacrifice a lot in either speed or flexibility. The Tensorpack API brings speed and flexibility together.

Tensorpack is Yet Another TF high-level API, but different in:

• Focus on training speed.

  • Speed comes for free with tensorpack – it uses TensorFlow in the efficient way with no extra overhead. On common CNNs, it runs 1.2~5x faster than the equivalent Keras code.

  • Data-parallel multi-GPU/distributed training strategy is off-the-shelf to use. It scales as well as Google’s official benchmark. You cannot beat its speed unless you’re a TensorFlow expert.

  • See tensorpack/benchmarks for some benchmark scripts.

• Focus on large datasets.

  • You don’t usually need tf.data. Symbolic programming often makes data processing harder. Tensorpack helps you efficiently process large datasets (e.g. ImageNet) in pure Python with autoparallelization.

• It’s not a model wrapper.

  • There are already too many symbolic function wrappers in the world. Tensorpack includes only a few common models, but you can use any symbolic function library inside tensorpack, including tf.layers/Keras/slim/tflearn/tensorlayer/…

See Tutorials to know more about these features:

• Tutorials
  • Introduction
    • What is tensorpack?
    • Why tensorpack?
    • A High Level Glance
  • Basic Tutorials
    • Trainers
    • Training Interface
    • Callbacks
    • Symbolic Layers
    • Save and Load models
    • Summary and Logging
    • Inference
    • FAQs
  • DataFlow Tutorials
    • DataFlow
    • Why DataFlow?
    • Write a DataFlow
    • Parallel DataFlow
    • Efficient DataFlow
  • Advanced Tutorials
    • Input Pipeline
    • Write a Callback
    • Write an Image Augmentor
    • Write an Augmentor
    • Write a Layer
    • Understand Trainer
    • Performance Tuning
• API Documentation
  • tensorpack.dataflow package
  • tensorpack.dataflow.dataset package
  • tensorpack.dataflow.imgaug package
  • tensorpack.input_source package
  • tensorpack.models package
  • tensorpack.callbacks package
  • tensorpack.graph_builder package
  • tensorpack.train package
  • tensorpack.predict package
  • tensorpack.tfutils package
    • tensorpack.tfutils.argscope module
    • tensorpack.tfutils.collection module
    • tensorpack.tfutils.gradproc module
    • tensorpack.tfutils.tower module
    • tensorpack.tfutils.scope_utils module
    • tensorpack.tfutils.optimizer module
    • tensorpack.tfutils.sesscreate module
    • tensorpack.tfutils.sessinit module
    • tensorpack.tfutils.summary module
    • tensorpack.tfutils.varmanip module
    • tensorpack.tfutils.varreplace module
    • tensorpack.tfutils.export module
    • tensorpack.tfutils.dependency module
    • Other functions in tensorpack.tfutils module
  • tensorpack.utils package
    • tensorpack.utils.argtools module
    • tensorpack.utils.concurrency module
    • tensorpack.utils.fs module
    • tensorpack.utils.loadcaffe module
    • tensorpack.utils.logger module
    • tensorpack.utils.serialize module
    • tensorpack.utils.stats module
    • tensorpack.utils.timer module
    • tensorpack.utils.viz module
    • tensorpack.utils.gpu module
  • tensorpack.contrib package