Torch Toolbox

A collection of snippets and libraries for Torch from Purdue e-Lab.

• Adversarial to generate adversarial examples using fast gradient sign method
• Bincompat to read and write Torch7's 64-bit file format on 32-bit platforms
• BN-absorber to merge batch normalization module into convolution layer
• Convert-MM to replace the ConvolutionMM with Convolution to support stride on CPU
• Face-align-Cao to find facial landmarks and align the face using regression model
• Face-align-Kazemi to find facial landmarks and align the face using regression trees
• Dataset-tools to augment folder of images
• GPU-RAM to monitor the GPU's memory usage
• GUI provides some examples for playing with the Qt Graphical User Interface
• Net-toolkit is a Lua package for smart/light network saving
• Re-plotting to recover bad plot and merging splitted training
• Sanitize to free temporary buffers in order to reduce a model size before saving on disk
• Try-model to load a new model's architecture and visualise its memory consumption
• Tutorials to help you set up Torch-related environment
  • Setup an Intel Compute Stick
  • Setup an Ubuntu GPU box
  • Recover a Jetson TK1 FS
  • Update a new Jetson TK1
  • Install CUDA on TK1
  • Install CUDNN on TK1
• Video-decoder to receive and decode video from various sources
• Video-transmitter to encode and transmit video on the fly
• Weight-init to reset weight and bias to have desired statistics
• demo-core core scripts for the demo-visor
• demo-visor to process a video or images with a network

License

MIT