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
- 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