"regularizeCNN" provides essential tools for evaluating the effects of different regularizers/optimizers when training a neural network.
- i7-950 or equivalent
- 8G+ DRAM
- Nvidia Geforce GPU with CUDA capability. 8G graphics memory.
- i7-950 or equivalent
- 8G+ DRAM
** Listed is my working environment. Some other may also work.
- Win7/8/10 /Ubuntu 18.04+/ Debian 9.7.0+
- Python 3.7
- Pytorch 1.5.0 + torchvision 0.6.0
- numpy 1.18.5
- matplotlib 3.2.2
- tqdm 4.46.1
- pillow 7.2.0
demoTrain.py : Train a new neural-net with specified dataset and save to file "._outputs/{timestamp}.model".
demoAdversary.py : Load a pre-trained net from file and generate adversary samples that can fool the net.