# Copyright (c) 2015-present, Facebook, Inc. # All rights reserved. import argparse import datetime import numpy as np import time import torch import torch.backends.cudnn as cudnn import json from pathlib import Path from torch.cuda.amp import GradScaler, autocast from torchvision.utils import make_grid import importlib import logging import torch.nn as nn import sys from typing import Iterable, Optional import logging import torch.distributed as dist import os from torch.utils.data import DataLoader from torch.utils.tensorboard import SummaryWriter from PIL import Image, ImageDraw, ImageFont from tqdm import tqdm import matplotlib.pyplot as plt from utils.util import show_params, visualize_features, init_distributed_mode, get_rank, get_world_size from dataset.build_dataset import CustomCocoDataset from omegaconf import OmegaConf from model import ControlLDM, Diffusion from utils.common import instantiate_from_config from utils.sampler import SpacedSampler def get_args_parser(): parser = argparse.ArgumentParser('DeiT training and evaluation script', add_help=False) parser.add_argument('--batch-size', default=64, type=int) parser.add_argument('--epochs', default=300, type=int) parser.add_argument('--bce-loss', action='store_true') parser.add_argument('--unscale-lr', action='store_true') # Feature extractor parameters parser.add_argument('--model', default='deit_base_patch16_224', type=str) parser.add_argument('--target_model', default='deit_base_patch16_224', type=str) parser.add_argument('--input-size', default=224, type=int, help='images input size') parser.add_argument('--drop', type=float, default=0.0, metavar='PCT', help='Dropout rate (default: 0.)') parser.add_argument('--drop-path', type=float, default=0.1, metavar='PCT', help='Drop path rate (default: 0.1)') parser.add_argument('--model-ema', action='store_true') parser.add_argument('--no-model-ema', action='store_false', dest='model_ema') parser.set_defaults(model_ema=True) parser.add_argument('--model-ema-decay', type=float, default=0.99996, help='') parser.add_argument('--model-ema-force-cpu', action='store_true', default=False, help='') # CLDM parameters parser.add_argument("--config", type=str, required=True) parser.add_argument('--total_train_steps', default=300000, type=int) parser.add_argument('--resume_path', default='', help='resume from checkpoint for controlnet') parser.add_argument('--cldm_learning_rate', default=1e-4, type=float, help='learning rate for controlnet') parser.add_argument('--sd_locked', default=True, help='whether to lock the sd of controlnet') parser.add_argument('--only_mid_control', default=False, help='only control the middle layers of controlnet') parser.add_argument('--cldm_yaml', default='./models/cldm_v15.yaml', help='yaml file for controlnet') parser.add_argument('--exp_dir', default='./exp', help='experiment directory') parser.add_argument('--image_floder', default='./data', help='training image floder') parser.add_argument("--log_every", default=20, type=int, help="log every n steps") parser.add_argument("--ckpt_every", default=1000, type=int, help="save checkpoint every n steps") parser.add_argument("--image_every", default=1000, type=int, help="log image every n steps") # parser.add_argument('--controlnet_dir', default='', type=str, help='resume from checkpoint for controlnet') parser.add_argument('--global_step', default=0, type=int, help='global step') parser.add_argument('--ddim_steps', default=50, type=int, help='ddim steps') parser.add_argument('--eta', default=0, type=float, help='ddim eta') parser.add_argument('--unconditional_guidance_scale', default=1.25, type=float, help='unconditional guidance scale') # Optimizer parameters parser.add_argument('--opt', default='adamw', type=str, metavar='OPTIMIZER', help='Optimizer (default: "adamw"') parser.add_argument('--opt-eps', default=1e-8, type=float, metavar='EPSILON', help='Optimizer Epsilon (default: 1e-8)') parser.add_argument('--opt-betas', default=None, type=float, nargs='+', metavar='BETA', help='Optimizer Betas (default: None, use opt default)') parser.add_argument('--clip-grad', type=float, default=None, metavar='NORM', help='Clip gradient norm (default: None, no clipping)') parser.add_argument('--momentum', type=float, default=0.9, metavar='M', help='SGD momentum (default: 0.9)') parser.add_argument('--weight-decay', type=float, default=0.05, help='weight decay (default: 0.05)') # Learning rate schedule parameters parser.add_argument('--sched', default='cosine', type=str, metavar='SCHEDULER', help='LR scheduler (default: "cosine"') parser.add_argument('--lr', type=float, default=4e-4, metavar='LR', help='learning rate (default: 5e-4)') parser.add_argument('--lr-noise', type=float, nargs='+', default=None, metavar='pct, pct', help='learning rate noise on/off epoch percentages') parser.add_argument('--lr-noise-pct', type=float, default=0.67, metavar='PERCENT', help='learning rate noise limit percent (default: 0.67)') parser.add_argument('--lr-noise-std', type=float, default=1.0, metavar='STDDEV', help='learning rate noise std-dev (default: 1.0)') parser.add_argument('--warmup-lr', type=float, default=1e-6, metavar='LR', help='warmup learning rate (default: 1e-6)') parser.add_argument('--min-lr', type=float, default=1e-5, metavar='LR', help='lower lr bound for cyclic schedulers that hit 0 (1e-5)') parser.add_argument('--decay-epochs', type=float, default=30, metavar='N', help='epoch interval to decay LR') parser.add_argument('--warmup-epochs', type=int, default=5, metavar='N', help='epochs to warmup LR, if scheduler supports') parser.add_argument('--cooldown-epochs', type=int, default=10, metavar='N', help='epochs to cooldown LR at min_lr, after cyclic schedule ends') parser.add_argument('--patience-epochs', type=int, default=10, metavar='N', help='patience epochs for Plateau LR scheduler (default: 10') parser.add_argument('--decay-rate', '--dr', type=float, default=0.1, metavar='RATE', help='LR decay rate (default: 0.1)') # Augmentation parameters parser.add_argument('--color-jitter', type=float, default=0.3, metavar='PCT', help='Color jitter factor (default: 0.3)') parser.add_argument('--aa', type=str, default='rand-m9-mstd0.5-inc1', metavar='NAME', help='Use AutoAugment policy. "v0" or "original". " + \ "(default: rand-m9-mstd0.5-inc1)'), parser.add_argument('--smoothing', type=float, default=0.1, help='Label smoothing (default: 0.1)') parser.add_argument('--train-interpolation', type=str, default='bicubic', help='Training interpolation (random, bilinear, bicubic default: "bicubic")') parser.add_argument('--repeated-aug', action='store_true') parser.add_argument('--no-repeated-aug', action='store_false', dest='repeated_aug') parser.set_defaults(repeated_aug=True) parser.add_argument('--train-mode', action='store_true') parser.add_argument('--no-train-mode', action='store_false', dest='train_mode') parser.set_defaults(train_mode=True) parser.add_argument('--ThreeAugment', action='store_true') #3augment parser.add_argument('--src', action='store_true') #simple random crop # add dataset parameters parser.add_argument('--global_crops_size', '--img_size', default=224, type=int, help="this should be equal to image size") parser.add_argument('--patch_size', default=16, type=int, help="patch size for vit patch embedding") # add masking parameter parser.add_argument('--mask_ratio', default=(0.1, 0.5), type=float, nargs='+', help="mask ratio can be either a value or a range") parser.add_argument('--mask_probability', default=0., type=float, help="how many samples with be applied with masking") parser.add_argument('--mask_first_n', action='store_true', help="mask the first n sample to avoid shuffling. Needed for MAE-style encoder") parser.add_argument('--clone_batch', default=1, type=int, help="how many times to clone the batch for masking (default: 1, not cloning)") # * Random Erase params parser.add_argument('--reprob', type=float, default=0.25, metavar='PCT', help='Random erase prob (default: 0.25)') parser.add_argument('--remode', type=str, default='pixel', help='Random erase mode (default: "pixel")') parser.add_argument('--recount', type=int, default=1, help='Random erase count (default: 1)') parser.add_argument('--resplit', action='store_true', default=False, help='Do not random erase first (clean) augmentation split') # * Mixup params parser.add_argument('--mixup', type=float, default=0.8, help='mixup alpha, mixup enabled if > 0. (default: 0.8)') parser.add_argument('--cutmix', type=float, default=1.0, help='cutmix alpha, cutmix enabled if > 0. (default: 1.0)') parser.add_argument('--cutmix-minmax', type=float, nargs='+', default=None, help='cutmix min/max ratio, overrides alpha and enables cutmix if set (default: None)') parser.add_argument('--mixup-prob', type=float, default=1.0, help='Probability of performing mixup or cutmix when either/both is enabled') parser.add_argument('--mixup-switch-prob', type=float, default=0.5, help='Probability of switching to cutmix when both mixup and cutmix enabled') parser.add_argument('--mixup-mode', type=str, default='batch', help='How to apply mixup/cutmix params. Per "batch", "pair", or "elem"') # Distillation parameters parser.add_argument('--teacher-model', default='base', type=str) parser.add_argument('--teacher-path', type=str, default='') parser.add_argument('--distillation-type', default='none', choices=['none', 'soft', 'hard'], type=str, help="") parser.add_argument('--distillation-alpha', default=0.5, type=float, help="") parser.add_argument('--distillation-tau', default=1.0, type=float, help="") parser.add_argument('--lambda_token', type=float, default=1.0) parser.add_argument('--lambda_fea', type=float, default=1.0) parser.add_argument('--lambda_patch', type=float, default=1.0) # * Cosub params parser.add_argument('--cosub', action='store_true') # * Finetuning params parser.add_argument('--finetune', default='', help='finetune from checkpoint') parser.add_argument('--attn-only', action='store_true') parser.add_argument('--weight_inherit', default='') # Dataset parameters parser.add_argument('--data-path', default='/datasets01/imagenet_full_size/061417/', type=str, help='dataset path') parser.add_argument('--data-set', default='IMNET', choices=['CIFAR', 'IMNET', 'IMNET_ibot', 'IMNET_ibot_aug', 'IMNET_ibot_fast_aug', 'INAT', 'INAT19', 'IMNET_L', 'IMNET_L_ibot'], type=str, help='Image Net dataset path') parser.add_argument('--inat-category', default='name', choices=['kingdom', 'phylum', 'class', 'order', 'supercategory', 'family', 'genus', 'name'], type=str, help='semantic granularity') parser.add_argument('--output_dir', default='', help='path where to save, empty for no saving') parser.add_argument('--log_dir', default='/data1/qiyp/Proteus-pytorch/pretrain/log/DINOv2_training/log', type=str, help='saving logging info every 20 iters') parser.add_argument('--device', default='cuda', help='device to use for training / testing') parser.add_argument('--seed', default=231, type=int) parser.add_argument('--start_epoch', default=0, type=int, metavar='N', help='start epoch') parser.add_argument('--eval', action='store_true', help='Perform evaluation only') parser.add_argument('--eval-crop-ratio', default=0.875, type=float, help="Crop ratio for evaluation") parser.add_argument('--dist-eval', action='store_true', default=False, help='Enabling distributed evaluation') parser.add_argument('--num_workers', default=8, type=int) parser.add_argument('--pin-mem', action='store_true', help='Pin CPU memory in DataLoader for more efficient (sometimes) transfer to GPU.') parser.add_argument('--no-pin-mem', action='store_false', dest='pin_mem', help='') parser.set_defaults(pin_mem=True) # distributed training parameters parser.add_argument('--distributed', action='store_true', default=False, help='Enabling distributed training') parser.add_argument('--world_size', default=1, type=int, help='number of distributed processes') parser.add_argument('--dist_url', default='env://', help='url used to set up distributed training') return parser def setup_logger(log_dir, rank=0): if rank != 0: return # 只有主进程（rank 0）配置日志记录器 log_formatter = logging.Formatter("%(asctime)s [%(levelname)-5.5s] %(message)s") root_logger = logging.getLogger() root_logger.setLevel(logging.INFO) log_file_handler = logging.FileHandler(log_dir, encoding='utf-8') log_file_handler.setFormatter(log_formatter) root_logger.addHandler(log_file_handler) log_stream_handler = logging.StreamHandler(sys.stdout) log_stream_handler.setFormatter(log_formatter) root_logger.addHandler(log_stream_handler) logging.info('Logging file is %s' % log_dir) def main(args): init_distributed_mode(args) print(args) device = torch.device(args.device) # 获取当前进程的 rank rank = dist.get_rank() if dist.is_initialized() else 0 # set up logger os.makedirs(args.log_dir, exist_ok=True) setup_logger(args.log_dir + '/' + time.strftime('%Y%m%d_%H%M%S') + '.log', rank) logging.info('job dir: {}'.format(os.path.dirname(os.path.realpath(__file__)))) logging.info("{}".format(args).replace(', ', ',\n') + '\n') # set_seed(231) # fix the seed for reproducibility seed = args.seed + get_rank() torch.manual_seed(seed) np.random.seed(seed) # random.seed(seed) cudnn.benchmark = True # Setup an experiment folder exp_dir = args.exp_dir os.makedirs(exp_dir, exist_ok=True) ckpt_dir = os.path.join(exp_dir, "checkpoints") os.makedirs(ckpt_dir, exist_ok=True) logging.info(f"Experiment directory created at {exp_dir}") # Create Dataset img_folder = args.image_floder dataset = CustomCocoDataset(img_folder) logging.info(f"Loaded train dataset with {len(dataset)} samples") # Distributed Sampler if args.distributed: num_tasks = get_world_size() global_rank = get_rank() sampler_train = torch.utils.data.DistributedSampler( dataset, num_replicas=num_tasks, rank=global_rank, shuffle=True ) else: sampler_train = torch.utils.data.RandomSampler(dataset) logging.info("Sampler_train = %s" % str(sampler_train)) # Dataloader dataloader = DataLoader( dataset, sampler=sampler_train, batch_size=args.batch_size, num_workers=args.num_workers, pin_memory=args.pin_mem, drop_last=True) logging.info('Dataloader created') # ========== create IF model ========== # logging.info(f"# ===== Creating Feature Extractor: {args.model} ===== #") meta_arch_module = importlib.import_module(args.model) # models_proteus_dinov2 MetaArch = meta_arch_module.MetaArch model = MetaArch(args) logging.info("Model = %s" % str(model)) if args.finetune: checkpoint = torch.load(args.finetune, map_location='cpu') if 'state_dict' in checkpoint: pretrained_dict = checkpoint['state_dict'] elif 'model' in checkpoint: pretrained_dict = checkpoint['model'] else: pretrained_dict = checkpoint missing_keys, unexpected_keys = model.load_state_dict(pretrained_dict, False) logging.info('Finetuning from %s' % args.finetune) logging.info('missing_keys: %s' % str(missing_keys)) logging.info('unexpected_keys: %s' % str(unexpected_keys)) if args.attn_only: for name_p,p in model.named_parameters(): if '.attn.' in name_p: p.requires_grad = True else: p.requires_grad = False try: model.head.weight.requires_grad = True model.head.bias.requires_grad = True except: model.fc.weight.requires_grad = True model.fc.bias.requires_grad = True try: model.pos_embed.requires_grad = True except: print('no position encoding') try: for p in model.patch_embed.parameters(): p.requires_grad = False except: print('no patch embed') model.to(device) model.eval() # ========= create cldm ========= # cfg = OmegaConf.load(args.config) cldm: ControlLDM = instantiate_from_config(cfg.model.cldm) sd = torch.load(cfg.train.sd_path, map_location="cpu")["state_dict"] unused = cldm.load_pretrained_sd(sd) logging.info(f"strictly load pretrained SD weight from {cfg.train.sd_path}\n" f"unused weights: {unused}") if cfg.train.resume: ckpt = torch.load(cfg.train.resume, map_location="cpu") cldm_state_dict = ckpt['model_state_dict'] cldm.load_controlnet_from_ckpt(cldm_state_dict) logging.info(f"strictly load controlnet weight from checkpoint: {cfg.train.resume}") if 'global_step' in ckpt: global_step = ckpt['global_step'] if 'epoch' in ckpt: epoch = ckpt['epoch'] logging.info(f"Resumed from global step {global_step}, epoch {epoch}") else: init_with_new_zero, init_with_scratch = cldm.load_controlnet_from_unet() logging.info(f"strictly load controlnet weight from pretrained SD\n" f"weights initialized with newly added zeros: {init_with_new_zero}\n" f"weights initialized from scratch: {init_with_scratch}") global_step = 0 epoch = 0 cldm = cldm.to(device) if args.distributed: cldm = torch.nn.parallel.DistributedDataParallel(cldm, device_ids=[args.gpu], find_unused_parameters=True) cldm_without_ddp = cldm.module else: cldm_without_ddp = cldm diffusion: Diffusion = instantiate_from_config(cfg.model.diffusion) # Setup optimizer: optimizer = torch.optim.AdamW(cldm_without_ddp.controlnet.parameters(), lr=args.cldm_learning_rate) if cfg.train.resume: checkpoint = torch.load(cfg.train.resume, map_location="cpu") # 恢复优化器状态 if 'optimizer_state_dict' in checkpoint: optimizer.load_state_dict(checkpoint['optimizer_state_dict']) logging.info(f"Optimizer state loaded from checkpoint: {cfg.train.resume}") show_params(cldm_without_ddp.controlnet) ddim_sampler = SpacedSampler(diffusion.betas) logging.info("# ========== Building model done! ========== #") # Preparation for training: cldm_without_ddp.train() diffusion.to(device) # 使用 GradScaler 进行混合精度训练 scaler = GradScaler(enabled=True) if cfg.train.resume: checkpoint = torch.load(cfg.train.resume, map_location="cpu") # 恢复优化器状态 if 'scaler_state_dict' in checkpoint: scaler.load_state_dict(checkpoint['scaler_state_dict']) # Load GradScaler state logging.info(f"Loss Scaler state loaded from checkpoint: {cfg.train.resume}") max_steps = args.total_train_steps step_loss = [] epoch_loss = [] # Create SummaryWriter only in the main process if rank == 0: writer = SummaryWriter(exp_dir) logging.info(f"Training for {max_steps} steps...") while global_step < max_steps: pbar = tqdm(iterable=dataloader, unit="batch") for batch in dataloader: if args.distributed: sampler_train.set_epoch(epoch) gt = batch["jpg"].to(device) hint = batch["hint"].to(device) prompt = batch["txt"] _, _, h, w = hint.shape _, _, H, W = gt.shape with torch.no_grad(): with autocast(dtype=torch.bfloat16): # 使用 bf16 自动混合精度 # z_0 = cldm_without_ddp.get_latent(2 * gt - 1) # rescale to [-1, 1] z_0 = cldm_without_ddp.vae_encode(2 * gt - 1) features_dict = model.student.backbone(hint, is_training=True) features = features_dict['x_norm_patchtokens'] features, _ = model.info_bottleneck(features, is_training=False) features = features.view(-1, h//14, w//14, features.shape[2]) # [B, h, w, c] features = features.permute(0, 3, 1, 2) features = (features - features.mean()) / features.std() features = torch.clamp(features, -5, 5) features_not_zero = features if torch.rand(1).item() < 0.1: # 10% of the time, set features to zero features = torch.zeros_like(features) cond = cldm_without_ddp.prepare_condition(features, prompt) # cond["c_txt"]: txt embed, cond["c_img"]: features with autocast(dtype=torch.bfloat16): # bf16 自动混合精度 # t = torch.randint(0, cldm_without_ddp.num_timesteps, (z_0.shape[0],), device=device).long() t = torch.randint(0, diffusion.num_timesteps, (z_0.shape[0],), device=device).long() # loss, _ = cldm_without_ddp.p_losses(z_0, cond, t) loss = diffusion.p_losses(cldm_without_ddp, z_0, t, cond) optimizer.zero_grad() # loss.backward() # optimizer.step() scaler.scale(loss).backward() # 使用 GradScaler 缩放损失 scaler.step(optimizer) scaler.update() # print("Gradient for input_hint_block:") # for name, param in cldm_without_ddp.controlnet.input_hint_block.named_parameters(): # if param.grad is not None: # print(f"{name} grad: {param.grad}") # else: # print(f"{name} grad is None") global_step += 1 step_loss.append(loss.item()) epoch_loss.append(loss.item()) pbar.update(1) pbar.set_description(f"Epoch: {epoch:04d}, Global Step: {global_step:07d}, Loss: {loss.item():.6f}") # Log loss values if rank == 0 and global_step % args.log_every == 0 and global_step > 0: avg_loss = np.mean(step_loss) step_loss.clear() writer.add_scalar("loss/loss_simple_step", avg_loss, global_step) # Save checkpoint if rank == 0 and global_step % args.ckpt_every == 0 and global_step > 0: # Create a checkpoint dictionary checkpoint = { 'model_state_dict': cldm_without_ddp.controlnet.state_dict(), 'optimizer_state_dict': optimizer.state_dict(), # Save optimizer state 'scaler_state_dict': scaler.state_dict(), # Save GradScaler state 'global_step': global_step, # Save current iteration 'epoch': epoch, # Save current epoch } ckpt_path = f"{ckpt_dir}/{global_step:07d}.pt" torch.save(checkpoint, ckpt_path) if rank == 0 and (global_step % args.image_every == 0 or global_step <= 1): eta = args.eta ddim_steps = args.ddim_steps cldm_without_ddp.eval() N = 1 log_features = features_not_zero[:N] log_cond = cldm_without_ddp.prepare_condition(log_features, prompt[:N]) log_uncond = cldm_without_ddp.prepare_condition(torch.zeros_like(log_features), prompt[:N]) log_gt = gt[:N] log_hint = hint[:N] # shape = (4, H // 8, W // 8) with torch.no_grad(): with torch.cuda.amp.autocast(dtype=torch.bfloat16): z = ddim_sampler.sample( model=cldm_without_ddp, device=device, steps=50, batch_size=len(log_gt), x_size=z_0.shape[1:], cond=log_cond, uncond=None, cfg_scale=1.0, x_T=None ) x_samples = cldm_without_ddp.vae_decode(z) x_samples = (x_samples + 1) / 2 z_cfg = ddim_sampler.sample( model=cldm_without_ddp, device=device, steps=50, batch_size=len(log_gt), x_size=z_0.shape[1:], cond=log_cond, uncond=log_uncond, cfg_scale=args.unconditional_guidance_scale, x_T=None ) x_samples_cfg = cldm_without_ddp.vae_decode(z_cfg) x_samples_cfg = (x_samples_cfg + 1) / 2 vis_features = visualize_features(log_features) for tag, image in [ ("image/samples", x_samples), ("image/samples_cfg", x_samples_cfg), ("image/gt", log_gt), ("image/condition", vis_features), ("image/hint", log_hint), ]: # Convert BFloat16 images to float32 before logging image = image.to(torch.float32) # Alternatively, you can convert to uint8 if required writer.add_image(tag, make_grid(image, nrow=1), global_step) cldm_without_ddp.train() if global_step == max_steps: break pbar.close() epoch += 1 if rank == 0: avg_epoch_loss = np.mean(epoch_loss) epoch_loss.clear() writer.add_scalar("loss/loss_simple_epoch", avg_epoch_loss, global_step) logging.info(f"Epoch: {epoch:04d}, Global Step: {global_step:07d}, Loss: {avg_epoch_loss:.6f}") logging.info("done!") if rank == 0: writer.close() if args.distributed: dist.barrier() if __name__ == '__main__': parser = argparse.ArgumentParser('DeiT training and evaluation script', parents=[get_args_parser()]) args = parser.parse_args() if args.output_dir: Path(args.output_dir).mkdir(parents=True, exist_ok=True) main(args)