Code for paper "Adversarial score matching and improved sampling for image generation"

Overview

Adversarial score matching and improved sampling for image generation

This repo contains the official implementation for the ICLR 2021 paper Adversarial score matching and improved sampling for image generation. It it a highly extended version of the original repo on score matching.

Discussion and more samples at https://ajolicoeur.wordpress.com/adversarial-score-matching-and-consistent-sampling.


Denoising score matching with Annealed Langevin Sampling (DSM-ALS) is a recent approach to generative modeling. Despite the convincing visual quality of samples, this method appears to perform worse than Generative Adversarial Networks (GANs) under the Frechet Inception Distance, a popular metric forgenerative models. We show that this apparent gap vanishes when denoising thefinal Langevin samples using the score network. In addition, we propose two improvements to DSM-ALS: 1) Consistent Annealed Sampling as a more stable alternative to Annealed Langevin Sampling, and 2) a hybrid training formulation, composed of both denoising score matching and adversarial objectives. By combining both of these techniques and exploring different network architectures, we elevate score matching methods and obtain results competitive with state-of-the-art image generation on CIFAR-10

Adversarial LSUN-Churches


Citation

If you find this code useful please cite us in your work:

@inproceedings{
  jolicoeur-martineau2021adversarial,
  title={Adversarial score matching and improved sampling for image generation},
  author={Alexia Jolicoeur-Martineau and R{\'e}mi Pich{\'e}-Taillefer and Ioannis Mitliagkas and Remi Tachet des Combes},
  booktitle={International Conference on Learning Representations},
  year={2021},
  url={https://openreview.net/forum?id=eLfqMl3z3lq}
}

Setup

Needed

Hyperparameter choice

  • Use "main.py --compute_approximate_sigma_max" to choose model.sigma_begin based on the current dataset (based on Technique 1 from https://arxiv.org/abs/2006.09011)
  • Use calculate_number_of_steps.R to choose model.num_classes (based on Technique 2 from https://arxiv.org/abs/2006.09011)
  • tune sampling.step_lr manually for consistent or non-consistent with n_sigma=1 (see Appendix B for how to extrapolate to n_sigma > 1 from the step_lr at n_sigma = 1)
  • Everything else can be left to default

To Replicate paper

To train a non-adversarial score network

python main.py --config cifar10_9999ema.yml --doc cifar10_bs128_L2_9999ema --ni

Log files will be saved in <exp>/logs/cifar10_bs128_L2_9999ema.

To train an adversarial score network

python main.py --config cifar10_9999ema.yml --doc cifar10_bs128_L2_9999ema_adam0_9_adamD-5_9_LSGAN_ --ni  --adam --adam_beta 0 .9 --D_adam --D_adam_beta -.5 .9 --adversarial

Log files will be saved in <exp>/logs/cifar10_bs128_L2_9999ema_adam0_9_adamD-5_9_LSGAN_.

To sample from a pre-trained score network (ex: cifar10_bs128_L2_9999ema, consistent, nsigma=1)

python main.py --sample --config cifar10_9999ema.yml -i cifar10_bs128_L2_9999ema --ni --consistent --nsigma 1 --step_lr 5.6e-6 --batch_size 100 --begin_ckpt 250000

Samples will be saved in <exp>/image_samples/cifar10_bs128_L2_9999ema.

To compute the FID for a range of checkpoints from a pre-trained score network (ex: cifar10_bs128_L2_9999ema, at 100k to 300k iterations)

python main.py --fast_fid --config cifar10_9999ema.yml -i cifar10_bs128_L2_9999ema --ni --consistent --nsigma 1 --step_lr 5.6e-6 --batch_size 4000 --fid_num_samples 10000 --begin_ckpt 100000 --end_ckpt 300000

FIDs will be saved in {args.fid_folder}/log_FID.txt.

Pretrained Score Network Checkpoints

Link: https://www.dropbox.com/s/dltiobdlsb2vhyo/DSM_ScoreNetwork_Pretrained.zip?dl=0

Download and unzip it to the exp folder.

FID statistics (for FID evaluation)

Link: https://www.dropbox.com/s/nhvp2tf1unxj08g/fid_stats.zip?dl=0

Download and unzip it to the exp/datasets folder.

Owner
Alexia Jolicoeur-Martineau
I'm a researcher in statistics and machine learning. I am particularly interested in GANs, denoising score-matching, and statistical divergences/metrics.
Alexia Jolicoeur-Martineau
Lambda-function - Python codes that allow notification of changes made to some services using the AWS Lambda Function

AWS Lambda Function This repository contains python codes that allow notificatio

Elif Apaydın 3 Feb 11, 2022
Disqus API bindings for Python

disqus-python Let's start with installing the API: pip install disqus-python Use the API by instantiating it, and then calling the method through dott

DISQUS 163 Oct 14, 2022
A Python Instagram Scraper for Downloading Profile's Posts, stories, ProfilePic and See the Details of Particular Instagram Profile.

✔ ✔ InstAstra ⚡ ⚡ ⁜ Description ~ A Python Instagram Scraper for Downloading Profile's Posts, stories, ProfilePic and See the Details of Particular In

12 Jun 23, 2022
A plugin for modmail-bot for stealing,making ,etc emojis

EmojiPlugin for the Modmail-bot My first plugin .. its very Basic I will make more and better too Only 3 commands for now emojiadd-supports .jpg, .png

1 Dec 28, 2021
Crosschat - A bot for cross-server communication

CrossChat A bot for cross-server communication. Running the bot To run the bot y

8 May 15, 2022
Telegram Bot to Filter posts in Bot Inline search

Inline-Filter-Bot A Telegram Bot for filter in Inline Features Unlimited Filters Supports all type of filters Supports Alert Button Using Common Marku

Code X Botz 67 Dec 26, 2022
AWS Workmail Migration Tool

WMigrate A tool for migrating AWS Workmail Users and Groups cross region and cross accounts. It also creates user and group aliases and adds the users

NK 1 Oct 27, 2021
Aws-lambda-requests-wrapper - Request/Response wrapper for AWS Lambda with API Gateway

AWS Lambda Requests Wrapper Request/Response wrapper for AWS Lambda with API Gat

1 May 20, 2022
A full-featured Python wrapper for the Onfleet API.

UPDATE: Please use Onfleet's wrapper instead. This repository is not maintained. https://github.com/onfleet/pyonfleet --- Python-Onfleet   python-onfl

Lionheart Software 11 Jan 13, 2022
With Google Drive API. My computer and my phone are in love now.

Channel trought Google Drive Google Drive API In this case, "Google Drive App" is the program. To install everything you need(has some extra things),

Luis Quiñones Requelme 1 Dec 15, 2021
LHXP・Official "LH - Cyber Security" Discord Leveling-Bot

LHXP・Official "LH - Cyber Security" Discord Leveling-Bot Based on nsde/NOVΛLIX Feature Overview /clear @user Requires admin permission Purges all XP

Felix・onlix 2 Mar 09, 2022
A python api to get info on covid-19

A python api to get info on covid-19

roof 2 Sep 18, 2022
3X Fast Telethon Based Bot

📺 YouTube Song Downloader Bot For Telegram 🔮 3X Fast Telethon Based Bot ⚜ Easy To Deploy 🤗

@Dk_king_offcial 1 Dec 09, 2021
KalmanFilterExercise - A Kalman Filter is a algorithmic filter that is used to estimate the state of an unknown variable

Kalman Filter Exercise What are Kalman Filters? A Kalman Filter is a algorithmic

4 Feb 26, 2022
🚀 A fast, flexible and lightweight Discord API wrapper for Python.

Krema A fast, flexible and lightweight Discord API wrapper for Python. Installation Unikorn unikorn add kremayard krema -no-confirmation Pip pip insta

Krema 20 Sep 04, 2022
Gorrabot is a bot made to automate checks and processes in the development process.

Gorrabot is a Gitlab bot made to automate checks and processes in the Faraday development. Features Check that the CHANGELOG is modified By default, m

Faraday 7 Dec 14, 2022
Python library for using SMS.ir web services

smsir smsir is a Python library for using SMS web services www.sms.ir Installation Use the package manager pip to install smsir. pip install smsir Usa

mohammad reza 2 Oct 14, 2022
Make low level API wrapper in fast, easy.

The lowrapper is a library for quickly and easily creating an environment for tapping the API without implementation.

tasuren 1 Oct 25, 2022
Ever wanted a dashboard for making your antispam? This is it.

Ever wanted a dashboard for making your antispam? This is it.

Skelmis 1 Oct 27, 2021
Seth Userbot with python

SETH-USERBOT DEPLOY TO HEROKU Group Support: String Session : Stay Support 🚀 ❁ LonamiWebs and Telethon © Credits ⚡ THANK YOU VERY MUCH FOR zeinzo Zei

seth 4 Jan 10, 2022