#!/bin/bash # Training export OPENAI_LOGDIR='OUTPUT/CELEBA-SDM-256CH' mpiexec -n 8 python image_train.py --data_dir ./data/celeba --dataset_mode celeba --lr 2e-5 --batch_size 4 --attention_resolutions 32,16,8 --diffusion_steps 1000 --image_size 256 --learn_sigma True \ --noise_schedule linear --num_channels 256 --num_head_channels 64 --num_res_blocks 2 --resblock_updown True --use_fp16 True --use_scale_shift_norm True --use_checkpoint True --num_classes 19 \ --class_cond True # Classifier-free Finetune export OPENAI_LOGDIR='OUTPUT/CELEBA-SDM-256CH-FINETUNE' mpiexec -n 8 python image_train.py --data_dir ./data/celeba --dataset_mode celeba --lr 2e-5 --batch_size 4 --attention_resolutions 32,16,8 --diffusion_steps 1000 --image_size 256 --learn_sigma True \ --noise_schedule linear --num_channels 256 --num_head_channels 64 --num_res_blocks 2 --resblock_updown True --use_fp16 True --use_scale_shift_norm True --use_checkpoint True --num_classes 19 \ --class_cond True --drop_rate 0.2 --resume_checkpoint OUTPUT/CELEBA-SDM-256CH/model.pt # Testing export OPENAI_LOGDIR='OUTPUT/CELEBA-SDM-256CH-TEST' mpiexec -n 8 python image_sample.py --data_dir ./data/celeba --dataset_mode celeba --attention_resolutions 32,16,8 --diffusion_steps 1000 --image_size 256 --learn_sigma True \ --noise_schedule linear --num_channels 256 --num_head_channels 64 --num_res_blocks 2 --resblock_updown True --use_fp16 True --use_scale_shift_norm True --num_classes 19 \ --class_cond True --batch_size 2 --num_samples 2000 --model_path OUTPUT/CELEBA-SDM-256CH-FINETUNE/ema_0.9999_best.pt --results_path RESULTS/CELEBA-SDM-256CH --s 1.5