txt2img_k.py

import argparse, os, sys, glob
import torch
import torch.nn as nn
import numpy as np
from omegaconf import OmegaConf
from PIL import Image
from tqdm import tqdm, trange
from itertools import islice
from einops import rearrange
from torchvision.utils import make_grid
import time
from pytorch_lightning import seed_everything
from torch import autocast
from contextlib import contextmanager, nullcontext
import accelerate

import k_samplers as K
from ldm.util import instantiate_from_config
from ldm.models.diffusion.ddim import DDIMSampler
from ldm.models.diffusion.plms import PLMSSampler


def chunk(it, size):
    it = iter(it)
    return iter(lambda: tuple(islice(it, size)), ())


def load_model_from_config(config, ckpt, verbose=False):
    print(f"Loading model from {ckpt}")
    pl_sd = torch.load(ckpt, map_location="cpu")
    if "global_step" in pl_sd:
        print(f"Global Step: {pl_sd['global_step']}")
    sd = pl_sd["state_dict"]
    model = instantiate_from_config(config.model)
    m, u = model.load_state_dict(sd, strict=False)
    if len(m) > 0 and verbose:
        print("missing keys:")
        print(m)
    if len(u) > 0 and verbose:
        print("unexpected keys:")
        print(u)

    model.cuda()
    model.eval()
    return model

class CFGDenoiser(nn.Module):
    def __init__(self, model):
        super().__init__()
        self.inner_model = model

    def forward(self, x, sigma, uncond, cond, cond_scale):
        x_in = torch.cat([x] * 2)
        sigma_in = torch.cat([sigma] * 2)
        cond_in = torch.cat([uncond, cond])
        uncond, cond = self.inner_model(x_in, sigma_in, cond=cond_in).chunk(2)
        return uncond + (cond - uncond) * cond_scale


def main():
    parser = argparse.ArgumentParser()

    parser.add_argument(
        "--prompt",
        type=str,
        nargs="?",
        default="a painting of a virus monster playing guitar",
        help="the prompt to render"
    )
    parser.add_argument(
        "--outdir",
        type=str,
        nargs="?",
        help="dir to write results to",
        default="out"
    )
    parser.add_argument(
        "--skip_grid",
        action='store_true',
        help="do not save a grid, only individual samples. Helpful when evaluating lots of samples",
    )
    parser.add_argument(
        "--skip_save",
        action='store_true',
        help="do not save individual samples. For speed measurements.",
    )
    parser.add_argument(
        "--ddim_steps",
        type=int,
        default=32,
        help="number of ddim sampling steps",
    )
    parser.add_argument(
        "--plms",
        action='store_true',
        help="use plms sampling",
    )
    parser.add_argument(
        "--laion400m",
        action='store_true',
        help="uses the LAION400M model",
    )
    parser.add_argument(
        "--fixed_code",
        action='store_true',
        help="if enabled, uses the same starting code across samples ",
    )
    parser.add_argument(
        "--ddim_eta",
        type=float,
        default=0.0,
        help="ddim eta (eta=0.0 corresponds to deterministic sampling",
    )
    parser.add_argument(
        "--n_iter",
        type=int,
        default=5,
        help="sample this often",
    )
    parser.add_argument(
        "--H",
        type=int,
        default=768,
        help="image height, in pixel space",
    )
    parser.add_argument(
        "--W",
        type=int,
        default=768,
        help="image width, in pixel space",
    )
    parser.add_argument(
        "--C",
        type=int,
        default=4,
        help="latent channels",
    )
    parser.add_argument(
        "--f",
        type=int,
        default=8,
        help="downsampling factor",
    )
    parser.add_argument(
        "--n_samples",
        type=int,
        default=1,
        help="how many samples to produce for each given prompt. A.k.a. batch size",
    )
    parser.add_argument(
        "--n_rows",
        type=int,
        default=0,
        help="rows in the grid (default: n_samples)",
    )
    parser.add_argument(
        "--scale",
        type=float,
        default=7.5,
        help="unconditional guidance scale: eps = eps(x, empty) + scale * (eps(x, cond) - eps(x, empty))",
    )
    parser.add_argument(
        "--from-file",
        type=str,
        help="if specified, load prompts from this file",
    )
    parser.add_argument(
        "--config",
        type=str,
        default="configs/stable-diffusion/v1-inference.yaml",
        help="path to config which constructs model",
    )
    parser.add_argument(
        "--ckpt",
        type=str,
        default="../mech-diffusion-v-1-A.ckpt",
        help="path to checkpoint of model",
    )
    parser.add_argument(
        "--seed",
        type=int,
        default=1337,
        help="the seed (for reproducible sampling)",
    )
    parser.add_argument(
        "--precision",
        type=str,
        help="evaluate at this precision",
        choices=["full", "autocast"],
        default="autocast"
    )
    opt = parser.parse_args()

    accelerator = accelerate.Accelerator()
    device = accelerator.device
    
    if opt.laion400m:
        print("Falling back to LAION 400M model...")
        opt.config = "configs/latent-diffusion/txt2img-1p4B-eval.yaml"
        opt.ckpt = "models/ldm/text2img-large/model.ckpt"
        opt.outdir = "outputs/txt2img-samples-laion400m"

    config = OmegaConf.load(f"{opt.config}")
    model = load_model_from_config(config, f"{opt.ckpt}")

    if opt.plms:
        sampler = PLMSSampler(model)
    else:
        sampler = DDIMSampler(model)
        
    model_wrap = K.external.CompVisDenoiser(model)
    sigma_min, sigma_max = model_wrap.sigmas[0].item(), model_wrap.sigmas[-1].item()

    os.makedirs(opt.outdir, exist_ok=True)
    outpath = opt.outdir

    batch_size = opt.n_samples
    n_rows = opt.n_rows if opt.n_rows > 0 else batch_size
    if not opt.from_file:
        prompt = opt.prompt
        assert prompt is not None
        data = [batch_size * [prompt]]

    else:
        print(f"reading prompts from {opt.from_file}")
        with open(opt.from_file, "r") as f:
            data = f.read().splitlines()
            data = list(chunk(data, batch_size))

    sample_path = os.path.join(outpath, "samples")
    os.makedirs(sample_path, exist_ok=True)
    base_count = len(os.listdir(sample_path))
    grid_count = len(os.listdir(outpath)) - 1

    start_code = None
    if opt.fixed_code:
        start_code = torch.randn([opt.n_samples, opt.C, opt.H // opt.f, opt.W // opt.f], device=device)

    precision_scope = autocast if opt.precision=="autocast" else nullcontext
    with torch.no_grad():
        with precision_scope("cuda"):
            with model.ema_scope():
                tic = time.time()
                all_samples = list()
                for n in trange(opt.n_iter, desc="Sampling", disable =not accelerator.is_main_process):
                    seed_everything(opt.seed+n)
                    seed = opt.seed+n
                    seeds = torch.randint(-2 ** 63, 2 ** 63 - 1, [accelerator.num_processes])
                    torch.manual_seed(seeds[accelerator.process_index].item())
                    for prompts in tqdm(data, desc="data", disable=not accelerator.is_main_process):
                        uc = None
                        if opt.scale != 1.0:
                            uc = model.get_learned_conditioning(batch_size * [""])
                        if isinstance(prompts, tuple):
                            prompts = list(prompts)
                        c = model.get_learned_conditioning(prompts)
                        shape = [opt.C, opt.H // opt.f, opt.W // opt.f]

                        sigmas = model_wrap.get_sigmas(opt.ddim_steps)
                        torch.manual_seed(opt.seed+n) # changes manual seeding procedure
                        x = torch.randn([opt.n_samples, *shape], device=device) * sigmas[0] # for GPU draw
                        model_wrap_cfg = CFGDenoiser(model_wrap)
                        extra_args = {'cond': c, 'uncond': uc, 'cond_scale': opt.scale}
                        samples_ddim = K.sampling.sample_euler_ancestral(model_wrap_cfg, x, sigmas, extra_args=extra_args, disable=not accelerator.is_main_process)
                        x_samples_ddim = model.decode_first_stage(samples_ddim)
                        x_samples_ddim = torch.clamp((x_samples_ddim + 1.0) / 2.0, min=0.0, max=1.0)
                        x_samples_ddim = accelerator.gather(x_samples_ddim)                        

                        if accelerator.is_main_process and not opt.skip_save:
                            for x_sample in x_samples_ddim:
                                x_sample = 255. * rearrange(x_sample.cpu().numpy(), 'c h w -> h w c')
                                Image.fromarray(x_sample.astype(np.uint8)).save(
                                    os.path.join(sample_path, f"{seed}_{base_count:05}.png"))
                                base_count += 1

                        if accelerator.is_main_process and not opt.skip_grid:
                            all_samples.append(x_samples_ddim)

                if accelerator.is_main_process and not opt.skip_grid:
                    # additionally, save as grid
                    grid = torch.stack(all_samples, 0)
                    grid = rearrange(grid, 'n b c h w -> (n b) c h w')
                    grid = make_grid(grid, nrow=n_rows)

                    # to image
                    grid = 255. * rearrange(grid, 'c h w -> h w c').cpu().numpy()
                    Image.fromarray(grid.astype(np.uint8)).save(os.path.join(outpath, f'grid-{grid_count:04}.png'))
                    grid_count += 1

                toc = time.time()

    print(f"Your samples are ready and waiting for you here: \n{outpath} \n"
          f" \nEnjoy.")


if __name__ == "__main__":
    main()