/csgm-robust-neurips

Primary LanguagePythonMIT LicenseMIT

Robust Compressed Sensing using Generative models

Setup

  1. Run git submodule init
  2. Run git submodule update
  3. Run python3.6 -m venv env
  4. Run source env/bin/activate
  5. Run pip install -r requirements.txt

Steps to reproduce the results

NOTE: Please run all commands from the root directory of the repository, i.e from csgm-robust-neurips/

Example code for running MOM tournaments:

python src/compressed_sensing.py --net pggan-256 --dataset range --input-type full-input --num-input-images 1 --batch-size 1 --measurement-type dense --adversarial-epsilon 0.02 --noise-std 0.01 --num-measurements 1000 --model-types mom --mom-batch-size 20 --optimizer-type yellowfin --learning-rate 1e-5 --momentum 0. --max-update-iter 2000 --num-random-restarts 2 --save-images --save-stats --checkpoint-iter 1 --cuda --gif --gif-dir temp --gif-iter 10