Guided-diffusion

This is the codebase for Diffusion Models Beat GANS on Image Synthesis.

Create the virtual env

conda create --name diffusion python=3.9
conda activate diffusion
pip install -r requirements.txt
conda install mpi4py

Finally

pip install -e .

Download pre-trained models

TBD

Testing models

sh test.sh 

To modify parameters in the test script

SAMPLE_FLAGS="--batch_size 2 --num_samples 20"

DDIM

For sampling with 250 step DDIM:

use `--timestep_respacing ddim250 --use_ddim True`

Training models

Run the training

sh train.sh exp1

Single GPU training

mpiexec -n 1 python scripts/image_train.py $TRAIN_FLAGS 

Multi-GPU training

mpiexec -n N python scripts/image_train.py $TRAIN_FLAGS

Training flags

TRAIN_FLAGS="--iterations 300000 --anneal_lr True --batch_size 4 --lr 3e-4 --save_interval 1000 --weight_decay 0.05"