• Search by Module
• Search by Words
• Search Projects

• Java
• Python
• JavaScript
• TypeScript
• C++
• Scala
• Blog

• benchmarks-master
  • .github
    • ISSUE_TEMPLATE.md
  • ImageNet
    • benchmark-opencv-resize.py
    • benchmark-tfdata.py
    • benchmark-dataflow.py
    • dump-lmdb.py
    • augmentors.py
    • symbolic_imagenet.py
    • README.md
  • LICENSE
  • DCGAN
    • README.md
  • profile-import
    • profile-import.py
    • README.md
    • import_profiler.py
  • Cifar10-fast
    • cifar10-fast.py
    • README.md
  • MaskRCNN
    • README.md
    • maskrcnn.patch
  • README.md
  • other-wrappers
    • tflearn.vgg.py
    • tensorpack.cifar10.py
    • tensorpack.alexnet.py
    • tensorpack.vgg.py
    • keras.vgg.py
    • keras.resnet.py
    • tflearn.cifar10.py
    • keras.alexnet.py
    • keras.cifar10.py
    • README.md
    • tensorpack.resnet.py
  • ResNet-Horovod
    • imagenet-resnet-horovod.py
    • resnet_model.py
    • imagenet_utils.py
    • README.md
    • serve-data.py
    • slurm.script
  • ResNet-MultiGPU
    • tfbench
      • convnet_builder.py
      • resnet_model.py
      • model.py
      • model_config.py
      • __init__.py
    • resnet-multigpu.py
    • README.md
  • .gitignore
  • tox.ini

tensorpack benchmarks

We use TensorFlow in the efficient way. Tensorpack is:

• As fast as tensorflow/benchmarks in multi-GPU ResNet training
• 1.2x~5x faster than Keras & tflearn in common CNNs
• Able to reproduce "ImageNet in one hour" with 256 GPUs
• Able to train Cifar10 to 94% accuracy within a minute
• 5x faster than matterport/Mask_RCNN
• 2.8x faster than DCGAN-tensorflow

All above claims can be reproduced with the corresponding code.