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
Final project in KAIST AI class

mmodal_mixer MLP-Mixer based Multi-modal image-text retrieval Image: Original image is cropped with 16 x 16 patch size without overlap. Then, it is re

SuperSuperMoon 5 May 30, 2022
pyForgeCert is a Python equivalent of the original ForgeCert written in C#.

pyForgeCert is a Python equivalent of the original ForgeCert written in C#.

Evi1cg 47 Oct 08, 2022
Simple calculator made in python

calculator Uma alculadora simples feita em python CMD, PowerShell, Bash ✔️ Início 💻 apt-get update apt-get upgrade -y apt-get install python git git

Spyware 8 Dec 28, 2021
C++ Environment InitiatorVisual Studio Code C / C++ Environment Initiator

Visual Studio Code C / C++ Environment Initiator Latest Version : v 1.0.1(2021/11/08) .exe link here About : Visual Studio Code에서 C/C++환경을 MinGW GCC/G

Junho Yoon 2 Dec 19, 2021
An educational platform for students

Watch N Learn About Watch N Learn is an educational platform for students. Watch N Learn incentivizes students to learn with fun activities and reward

Brian Law 3 May 04, 2022
Generating rent availability info from Effort rent

Rent-info Generating rent availability info from Effort rent Pre-Installation Latest version of python Pip module json, os, requests, datetime, time i

Laixuan 1 Oct 20, 2021
Live tracking, flight database and competition framework

SkyLines SkyLines is a web platform where pilots can share their flights with others after, or even during flight via live tracking. SkyLines is a sor

SkyLines 367 Dec 27, 2022
Pykeeb - A small Python script that prints out currently connected keyboards

pykeeb 🐍 ⌨️ A small Python script that detects and prints out currently connect

Jordan Duabe 1 May 08, 2022
This script is written with Python for selling steam community items automatically.

SteamCommunityItemAutoSell Description This script is written with Python for selling steam community items automatically. Install git clone https://g

14 Oct 26, 2022
program to store and update pokemons using SQL and Flask

Pokemon SQL and Flask Pokemons api in python. Technologies flask pymysql Description PokeCorp is a company that tracks pokemon and their trainers arou

Sara Hindy Salfer 1 Oct 20, 2021
Singularity Containers on Apple M1 (ARM64)

Singularity Containers on Apple M1 (ARM64) This is a repository containing a ready-to-use environment for singularity in arm64 (M1). It has been prepa

Manuel Parra 4 Nov 14, 2022
This is a library to do functional programming in Python.

Fpylib This is a library to do functional programming in Python. Index Fpylib Index Features Intelligents Ranges with irange Lazyness to functions Com

Fabián Vega Alcota 4 Jul 17, 2022
Reference management solution using Python and Notion.

notion-scholar Reference management solution using Python and Notion. The main idea of this app is to allow to furnish a Notion database using a BibTe

Thomas Hirtz 69 Dec 21, 2022
pydock - Docker-based environment manager for Python

pydock - Docker-based environment manager for Python ⚠️ pydock is still in beta mode, and very unstable. It is not recommended for anything serious. p

Alejandro Piad 16 Sep 18, 2021
Blender Addon for Snapping a UV to a specific part of a Tilemap

UVGridSnapper A simple Blender Addon for easier texturing. A menu in the UV editor allows a square UV to be snapped to an Atlas texture, or Tilemap. P

2 Jul 17, 2022
A script that will warn you, by opening a new browser tab, when there are new content in your favourite websites.

web check A script that will warn you, by opening a new browser tab, when there are new content in your favourite websites. What it does The script wi

Jaime Álvarez 52 Mar 15, 2022
Purge your likes and wall comments from VKontakte. Set yourself free from your digital footprint.

vk_liberator Regain liberty in the cruel social media world. This program assists you with purging your metadata from Russian social network VKontakte

20 Jun 11, 2021
Lock a program and kills it indefinitely if it is started.

Kill By Lock Lock a program and kills it indefinitely if it is started. How start it? It' simple, you just have to double-click on the python file (.p

1 Jan 12, 2022
Project Interface For nextcord-ext

Project Interface For nextcord-ext

nextcord-ext 1 Nov 13, 2021
use Notepad++ for real-time sync after python appending new log text

FTP远程log同步工具 使用Notepad++配合来获取实时更新的log文档效果 适用于FTP协议的log远程同步工具,配合MT管理器开启FTP服务器使用,通过Notepad++监听文本变化,更便捷的使用电脑查看方法注入打印后的信息 功能 过滤器 对每行要打印的文本使用回调函数筛选,支持链式调用

Liuhaixv 1 Oct 17, 2021