engine_fluid.py

import math
import sys
from typing import Iterable

import torch

import util.misc as misc
import util.lr_sched as lr_sched
from models.vae import DiagonalGaussianDistribution
import torch_fidelity
import shutil
import cv2
import numpy as np
import os
import copy
import time


def update_ema(target_params, source_params, rate=0.99):
    """
    Update target parameters to be closer to those of source parameters using
    an exponential moving average.

    :param target_params: the target parameter sequence.
    :param source_params: the source parameter sequence.
    :param rate: the EMA rate (closer to 1 means slower).
    """
    for targ, src in zip(target_params, source_params):
        targ.detach().mul_(rate).add_(src, alpha=1 - rate)


# def train_one_epoch(model, vae, t5_emb, model_params, ema_params,
#                     data_loader: Iterable, optimizer: torch.optim.Optimizer,
#                     device: torch.device, epoch: int, loss_scaler,
#                     log_writer=None, wandb=None, global_rank=None,
#                     args=None):
def train_one_epoch(model, vae, model_params, ema_params,
                    data_loader: Iterable, optimizer: torch.optim.Optimizer,
                    device: torch.device, epoch: int, loss_scaler,
                    log_writer=None, wandb=None, global_rank=None,
                    args=None):
    model.train(True)
    metric_logger = misc.MetricLogger(delimiter="  ")
    metric_logger.add_meter('lr', misc.SmoothedValue(window_size=1, fmt='{value:.6f}'))
    header = 'Epoch: [{}]'.format(epoch)
    print_freq = 20

    optimizer.zero_grad()

    if log_writer is not None:
        print('log_dir: {}'.format(log_writer.log_dir))

    for data_iter_step, (samples, text_emb) in enumerate(metric_logger.log_every(data_loader, print_freq, header)):
        interation_step = epoch * len(data_loader) + data_iter_step
        # we use a per iteration (instead of per epoch) lr scheduler
        lr_sched.adjust_learning_rate(optimizer, data_iter_step / len(data_loader) + epoch, args)

        samples = samples.to(device, non_blocking=True)

        text_emb = text_emb.to(device, non_blocking=True)
        # text_emb = t5_emb(text_emb)

        with torch.no_grad():
            if args.use_cached:
                moments = samples
                posterior = DiagonalGaussianDistribution(moments)
            else:
                posterior = vae.encode(samples).latent_dist
            
            if vae.config.shift_factor is not None:
                x = (posterior.sample() - vae.config.shift_factor) * vae.config.scaling_factor
            else:
                x = posterior.sample().mul_(vae.config.scaling_factor)
            # x = x.to(dtype)

        # forward
        with torch.cuda.amp.autocast():
            loss = model(x, text_emb, height=args.img_size, width=args.img_size)

        loss_value = loss.item()

        # Log metrics to WandB at the end of the epoch
        if wandb is not None:
            if global_rank == 0:
                wandb.log({
                    "steps": interation_step, # step
                    "loss": loss_value,
                    "learning_rate": optimizer.param_groups[0]["lr"]
                })

        if not math.isfinite(loss_value):
            print("Loss is {}, stopping training".format(loss_value))
            sys.exit(1)

        loss_scaler(loss, optimizer, clip_grad=args.grad_clip, parameters=model.parameters(), update_grad=True)
        optimizer.zero_grad()

        torch.cuda.synchronize()

        update_ema(ema_params, model_params, rate=args.ema_rate)

        metric_logger.update(loss=loss_value)

        lr = optimizer.param_groups[0]["lr"]
        metric_logger.update(lr=lr)

        loss_value_reduce = misc.all_reduce_mean(loss_value)
        if log_writer is not None:
            """ We use epoch_1000x as the x-axis in tensorboard.
            This calibrates different curves when batch size changes.
            """
            epoch_1000x = int((data_iter_step / len(data_loader) + epoch) * 1000)
            log_writer.add_scalar('train_loss', loss_value_reduce, epoch_1000x)
            log_writer.add_scalar('lr', lr, epoch_1000x)

    # gather the stats from all processes
    metric_logger.synchronize_between_processes()
    print("Averaged stats:", metric_logger)
    return {k: meter.global_avg for k, meter in metric_logger.meters.items()}

def cache_latents(vae, t5_emb, 
                  data_loader: Iterable,
                  device: torch.device,
                  args=None):
    metric_logger = misc.MetricLogger(delimiter="  ")
    header = 'Caching: '
    print_freq = 20

    for data_iter_step, (samples, prompts, paths) in enumerate(metric_logger.log_every(data_loader, print_freq, header)):

        samples = samples.to(device, non_blocking=True)

        with torch.no_grad():
            posterior = vae.encode(samples, return_dict=False)[0]
            moments = posterior.parameters
            posterior_flip = vae.encode(samples.flip(dims=[3]), return_dict=False)[0]
            moments_flip = posterior_flip.parameters
            text_emb = t5_emb(prompts)

        for i, path in enumerate(paths):
            save_path = os.path.join(args.cached_path, path + '.npz')
            os.makedirs(os.path.dirname(save_path), exist_ok=True)
            np.savez(save_path, moments=moments[i].cpu().numpy(), moments_flip=moments_flip[i].cpu().numpy())
            txt_path = os.path.join(args.txt_cached_path, path + '.npz')
            os.makedirs(os.path.dirname(txt_path), exist_ok=True)
            np.savez(txt_path, text_emb=text_emb[i].cpu().numpy())

        if misc.is_dist_avail_and_initialized():
            torch.cuda.synchronize()

    return