Release for Improved Denoising Diffusion Probabilistic Models

Overview

improved-diffusion

This is the codebase for Improved Denoising Diffusion Probabilistic Models.

Usage

This section of the README walks through how to train and sample from a model.

Installation

Clone this repository and navigate to it in your terminal. Then run:

pip install -e .

This should install the improved_diffusion python package that the scripts depend on.

Preparing Data

The training code reads images from a directory of image files. In the datasets folder, we have provided instructions/scripts for preparing these directories for ImageNet, LSUN bedrooms, and CIFAR-10.

For creating your own dataset, simply dump all of your images into a directory with ".jpg", ".jpeg", or ".png" extensions. If you wish to train a class-conditional model, name the files like "mylabel1_XXX.jpg", "mylabel2_YYY.jpg", etc., so that the data loader knows that "mylabel1" and "mylabel2" are the labels. Subdirectories will automatically be enumerated as well, so the images can be organized into a recursive structure (although the directory names will be ignored, and the underscore prefixes are used as names).

The images will automatically be scaled and center-cropped by the data-loading pipeline. Simply pass --data_dir path/to/images to the training script, and it will take care of the rest.

Training

To train your model, you should first decide some hyperparameters. We will split up our hyperparameters into three groups: model architecture, diffusion process, and training flags. Here are some reasonable defaults for a baseline:

MODEL_FLAGS="--image_size 64 --num_channels 128 --num_res_blocks 3"
DIFFUSION_FLAGS="--diffusion_steps 4000 --noise_schedule linear"
TRAIN_FLAGS="--lr 1e-4 --batch_size 128"

Here are some changes we experiment with, and how to set them in the flags:

  • Learned sigmas: add --learn_sigma True to MODEL_FLAGS
  • Cosine schedule: change --noise_schedule linear to --noise_schedule cosine
  • Reweighted VLB: add --use_kl True to DIFFUSION_FLAGS and add --schedule_sampler loss-second-moment to TRAIN_FLAGS.
  • Class-conditional: add --class_cond True to MODEL_FLAGS.

Once you have setup your hyper-parameters, you can run an experiment like so:

python scripts/image_train.py --data_dir path/to/images $MODEL_FLAGS $DIFFUSION_FLAGS $TRAIN_FLAGS

You may also want to train in a distributed manner. In this case, run the same command with mpiexec:

mpiexec -n $NUM_GPUS python scripts/image_train.py --data_dir path/to/images $MODEL_FLAGS $DIFFUSION_FLAGS $TRAIN_FLAGS

When training in a distributed manner, you must manually divide the --batch_size argument by the number of ranks. In lieu of distributed training, you may use --microbatch 16 (or --microbatch 1 in extreme memory-limited cases) to reduce memory usage.

The logs and saved models will be written to a logging directory determined by the OPENAI_LOGDIR environment variable. If it is not set, then a temporary directory will be created in /tmp.

Sampling

The above training script saves checkpoints to .pt files in the logging directory. These checkpoints will have names like ema_0.9999_200000.pt and model200000.pt. You will likely want to sample from the EMA models, since those produce much better samples.

Once you have a path to your model, you can generate a large batch of samples like so:

python scripts/image_sample.py --model_path /path/to/model.pt $MODEL_FLAGS $DIFFUSION_FLAGS

Again, this will save results to a logging directory. Samples are saved as a large npz file, where arr_0 in the file is a large batch of samples.

Just like for training, you can run image_sample.py through MPI to use multiple GPUs and machines.

You can change the number of sampling steps using the --timestep_respacing argument. For example, --timestep_respacing 250 uses 250 steps to sample. Passing --timestep_respacing ddim250 is similar, but uses the uniform stride from the DDIM paper rather than our stride.

To sample using DDIM, pass --use_ddim True.

Owner
OpenAI
OpenAI
Python MQTT v5.0 async client

gmqtt: Python async MQTT client implementation. Installation The latest stable version is available in the Python Package Index (PyPi) and can be inst

Gurtam 306 Jan 03, 2023
It really seems like Trump is trying to get his own social media started. Not a huge fan tbh.

FuckTruthSocial It really seems like Trump is trying to get his own social media started. Not a huge fan tbh. (When TruthSocial actually releases, I'l

0 Jul 18, 2022
A Python application that helps users determine their calorie intake, and automatically generates customized weekly meal and workout plans based on metrics computed using their physical parameters

A Python application that helps users determine their calorie intake, and automatically generates customized weekly meal and workout plans based on metrics computed using their physical parameters

Anam Iqbal 1 Jan 13, 2022
Pre-crisis Risk Management for Personal Finance

Антикризисный риск-менеджмент личных финансов Риск-менеджмент личных финансов условиях санкций и/или финансового кризиса: делаем сегодня все, чтобы за

Dmitry Petukhov 593 Jan 09, 2023
Lagrange Interpolation Method-Python

Lagrange Interpolation Method-Python The Lagrange interpolation formula is a way to find a polynomial, called Lagrange polynomial, that takes on certa

Motahare Soltani 2 Jul 05, 2022
tox-gh is a tox plugin which helps running tox on GitHub Actions with multiple different Python versions on multiple workers in parallel

tox-gh is a tox plugin which helps running tox on GitHub Actions with multiple different Python versions on multiple workers in parallel. This project is inspired by tox-travis.

tox development team 19 Dec 26, 2022
a url shortener with fastapi and tortoise-orm

fastapi-tortoise-orm-url-shortener a url shortener with fastapi and tortoise-orm

19 Aug 12, 2022
GWAS summary statistics files QC tool

SSrehab dependencies: python 3.8+ a GNU/Linux with bash v4 or 5. python packages in requirements.txt bcftools (only for prepare_dbSNPs) gz-sort (only

21 Nov 02, 2022
News-app - This is a news web app for reading news from different sources and topics

News-app - This is a news web app for reading news from different sources and topics

1 Feb 02, 2022
A simple tool made in Python language

Simple tool Uma simples ferramenta feita 100% em linguagem Python 💻 Requisitos: Python3 instalado em seu dispositivo Clonagem e acesso 📳 git clone h

josh washington 4 Dec 07, 2021
An example project which contains the Unity components necessary to complete Navigation2's SLAM tutorial with a Turtlebot3, using a custom Unity environment in place of Gazebo.

Navigation 2 SLAM Example This example provides a Unity Project and a colcon workspace that, when used together, allows a user to substitute Unity as

Unity Technologies 183 Jan 04, 2023
InverterApi - This project has been designed to take monitoring data from Voltronic, Axpert, Mppsolar PIP, Voltacon, Effekta

InverterApi - This project has been designed to take monitoring data from Voltronic, Axpert, Mppsolar PIP, Voltacon, Effekta

Josep Escobar 2 Sep 03, 2022
京东热爱狂欢趴&京东扫码获取cookie

京东热爱狂欢趴 一键完成任务脚本来袭 活动地址: https://wbbny.m.jd.com/babelDiy/Zeus/2s7hhSTbhMgxpGoa9JDnbDzJTaBB/index.html#/home 2021-06-02更新: 1、删除京东星推官 2、更新脚本,修复火爆问题 2021

xoyi 48 Dec 28, 2022
Git Hooks Tutorial.

Git Hooks Tutorial My public talk about this project at Sberloga: Git Hooks Is All You Need 1. Git Hooks 101 Init git repo: mkdir git_repo cd git_repo

Dani El-Ayyass 17 Oct 12, 2022
Tesla App Update Differences Extractor

Tesla App Update Differences Extractor Python program that finds the differences between two versions of the Tesla App. When Tesla updates the app a l

Adrian 5 Apr 11, 2022
Collections of python projects

nppy, mostly contains projects written in Python. Some projects are very simple while some are a bit lenghty and difficult(for beginners) Requirements

ghanteyyy 75 Dec 20, 2022
Desafio Final do Mod1 do Bootcamp EDC - v2 usando a RAIS

IGTI - Bootcamp Engenheiro de Dados Cloud Módulo 1 - Desafio Final - RAIS 2020 Código do Desafio Final V2 do Bootcamp Engenheiro de Dados Cloud do IGT

Neylson Crepalde 17 Nov 02, 2022
Sheet2export - FreeCAD macro to export spreadsheet

Description This is FreeCAD macro to export spreadsheet to file.

Darek L 3 Jul 09, 2022
Chicks get hostloc points regularly

hostloc_getPoints 小鸡定时获取hostloc积分 github action大规模失效,mjj平均一人10鸡,以下可以部署到自己的小鸡上

59 Dec 28, 2022
Rofi script to minimize / unminimize multiple windows in qtile

Qminimize Rofi script to minimize / unminimize multiple windows in qtile Additional requirements : EWMH module fuzzywuzzy module How to use it : - Clo

9 Sep 18, 2022