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
If you only have hash, you can still operate exchange

PTH Exchange If you only have hash, you can still operate exchange This project module is the same as my other project Exchange_SSRF, This project use

Jumbo 37 Dec 26, 2022
Converts a text file of songs to a playlist on your Spotify account.

Playlist Converter Convert a text file of songs to a playlist on your Spotify account. Create your playlists faster instead of manually searching for

Priya Aggarwal 18 Dec 21, 2022
Telegram Url Upload Bot With Same more Features ✨

Telegram Url Upload Bot With Same more Features ✨

Group Dc Bots 4 Feb 12, 2022
The best discord Nuk3r !

Discord - Nuker ☢️ Nuk3r ☢️ STEP 1 ✅ We go create discord bot ! [] Go on https://discord.com/developers/applications [] Set the name of your applica

2s.py 1 Apr 16, 2022
Visionary-OS: open source discord bot

Visionary-OS Our Visionary open source discord bot. Our goal is to create a discord bot, which is hosted by us, but every member of our community can

8 Jan 27, 2022
Replace sequence_IDs in gff3 based on given genome.fasta

gff-rename Replace the sequence IDs in a gff3 file with a set of provided sequence IDs from a genom.fasta. This is useful when a gff3 file is retrieve

tolkit 1 Nov 12, 2021
Pls give vaccine.

Pls Give Vaccine A script to spam yourself with vaccine notifications. Explore the docs » View Demo · Report Bug · Request Feature Table of Contents A

Rohan Mukherjee 3 Oct 27, 2021
The AWS Lambda Serverless Blind XSS App

Ass The AWS Lambda Serverless Blind XSS App 利用VPS配置XSS平台太麻烦了,如果利用AWS的Lambda那不就是一个域名的事情么?剩下的环境配置、HTTPS证书、隐私性、VPS续费都不用管了, 所以根据xless重写了Lambda平台的XSS,利用sla

cocokey 2 Dec 27, 2021
This is a script to export logs from AWS CloudTrail to a local file.

cloudtrail-export-logs This is a script to export logs from AWS CloudTrail to a local file. Getting Started Prerequisites python 3 boto3 pip Installin

Claick Assunção de Oliveira 2 Jan 02, 2022
Generate Heroku-like random names to use in your python applications

HaikunatorPY Generate Heroku-like random names to use in your python applications. Installation pip install haikunator Usage Haikunator is pretty sim

Atrox 116 Nov 15, 2022
Async wrapper over hentaichan.live

hentai-chan-api-async is a small asynchronous parser library that will allow you to easily use manga from https://hentaichan.live Recommended to use python3.7+

7 Dec 15, 2022
A Discord bot to play bluffing games like Dobbins or Bobbins

Usage: pip install -r requirements.txt python3 bot.py DISCORD_BOT_TOKEN Gameplay: All commands are case-insensitive, with trailing punctuation and spa

4 May 27, 2022
ShadowMusic - A Telegram Music Bot with proper functions written in Python with Pyrogram and Py-Tgcalls.

⭐️ Shadow Music ⭐️ A Telegram Music Bot written in Python using Pyrogram and Py-Tgcalls Ready to use method A Support Group, Updates Channel and ready

TeamShadow 8 Aug 17, 2022
Its The Basic Commands Of Termux

Its The Basic Commands Of Termux

ANKIT KUMAR 1 Dec 27, 2021
Simple Discord bot for snekbox (sandboxed Python code execution), self-host or use a global instance

snakeboxed Simple Discord bot for snekbox (sandboxed Python code execution), self-host or use a global instance

0 Jun 25, 2022
Retrieves GitHub Stats via `git_api` and flask.

GitHub User Search Created using Python3 and git_api, coded by JBYT27. About This is a project I decided to make for Kajam, but I decided to choose a

an aspirin 4 May 11, 2022
A tool that ensures consistent string quotes in your Python code.

pyquotes Single quotes are superior. And if you disagree, there's an option for this as well. In any case, quotes should be consistent throughout the

Adrian 9 Sep 13, 2022
Simple Reddit bot that replies to comments containing a certain word.

reddit-replier-bot Small comment reply bot based on PRAW. This script will scan the comments of a subreddit as they come in and look for a trigger wor

Kefendy 0 Jun 04, 2022
A module grouping multiple translation APIs

translatepy (originally: translate) An aggregation of multiple translation API Translate, transliterate, get the language of texts in no time with the

349 Jan 06, 2023