deepsense-ai / Keras-PyTorch-AvP-transfer-learning

We pit Keras and PyTorch against each other, showing their strengths and weaknesses in action. We present a real problem, a matter of life-and-death: distinguishing Aliens from Predators!

Keras vs. PyTorch: Alien vs. Predator recognition with transfer learning

Featured in deepsense.ai blog post Keras vs. PyTorch: Alien vs. Predator recognition with transfer learning, in which we discuss the differences. Code is in two Jupyter Notebooks:

• Transfer learning with ResNet-50 in Keras
• Transfer learning with ResNet-50 in PyTorch

See also the upcoming webinar (10 Oct 2018), in which we walk trough the code.

For plug&play interactive code, see the Neptune versions with fancy charts or these Kaggle Kernels:

• Transfer learning with ResNet-50 in Keras - Kaggle Kernel
• Transfer learning with ResNet-50 in PyTorch - Kaggle Kernel

Data

See also: Alien vs. Predator images | Kaggle. In general, there are 447 images for each class, split into two classes. Examples:

Requirements

If you want to run the code, see the requirements:

• Common:
  • jupyter==1.0.0
  • matplotlib==2.2.3
  • Pillow==5.2.0
  • h5py==2.8.0
• Keras:
  • tensorflow==1.10.1
  • Keras==2.2.2
• PyTorch:
  • torch==0.4.1
  • torchvision==0.2.1

Webinar info

• Link to the webinar (10 Oct 2018, 5 PM CEST)
• Slides from the webinar
• deepsense.ai's articles about Keras and PyTorch:
  • Keras or PyTorch as your first deep learning framework
  • Keras vs. PyTorch: Alien vs. Predator recognition with transfer learning
• Notebooks in Neptune
• Workshops:
  • open workshops at deepsense.ai's HQ
  • on-site workshops