YOLOv5 實作紀錄：從資料標註到工業相機偵測

# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
"""
在自定義數據集上訓練 YOLOv5 模型。
從最新的 YOLOv5 版本自動下載模型和數據集。

用法 - 單 GPU 訓練：
    $ python train.py --data coco128.yaml --weights yolov5s.pt --img 640  # from pretrained (recommended)
    $ python train.py --data coco128.yaml --weights '' --cfg yolov5s.yaml --img 640  # from scratch

用法 - 多 GPU DDP 訓練：
    $ python -m torch.distributed.run --nproc_per_node 4 --master_port 1 train.py --data coco128.yaml --weights yolov5s.pt --img 640 --device 0,1,2,3

模型:     https://github.com/ultralytics/yolov5/tree/master/models
數據集:   https://github.com/ultralytics/yolov5/tree/master/data
教程:   https://github.com/ultralytics/yolov5/wiki/Train-Custom-Data
"""

import argparse
import math
import os
import random
import sys
import time
from copy import deepcopy
from datetime import datetime
from pathlib import Path

import numpy as np
import torch
import torch.distributed as dist
import torch.nn as nn
import yaml
from torch.optim import lr_scheduler
from tqdm import tqdm

FILE = Path(__file__).resolve()
ROOT = FILE.parents[0]  # YOLOv5 root directory
if str(ROOT) not in sys.path:
    sys.path.append(str(ROOT))  # add ROOT to PATH
ROOT = Path(os.path.relpath(ROOT, Path.cwd()))  # relative

import val as validate  # for end-of-epoch mAP
from models.experimental import attempt_load
from models.yolo import Model
from utils.autoanchor import check_anchors
from utils.autobatch import check_train_batch_size
from utils.callbacks import Callbacks
from utils.dataloaders import create_dataloader
from utils.downloads import attempt_download, is_url
from utils.general import (LOGGER, TQDM_BAR_FORMAT, check_amp, check_dataset, check_file, check_git_info,
                           check_git_status, check_img_size, check_requirements, check_suffix, check_yaml, colorstr,
                           get_latest_run, increment_path, init_seeds, intersect_dicts, labels_to_class_weights,
                           labels_to_image_weights, methods, one_cycle, print_args, print_mutation, strip_optimizer,
                           yaml_save)
from utils.loggers import Loggers
from utils.loggers.comet.comet_utils import check_comet_resume
from utils.loss import ComputeLoss
from utils.metrics import fitness
from utils.plots import plot_evolve
from utils.torch_utils import (EarlyStopping, ModelEMA, de_parallel, select_device, smart_DDP, smart_optimizer,
                               smart_resume, torch_distributed_zero_first)

LOCAL_RANK = int(os.getenv('LOCAL_RANK', -1))  # https://pytorch.org/docs/stable/elastic/run.html
RANK = int(os.getenv('RANK', -1))
WORLD_SIZE = int(os.getenv('WORLD_SIZE', 1))
GIT_INFO = check_git_info()


def train(hyp, opt, device, callbacks):  # hyp is path/to/hyp.yaml or hyp dictionary
    save_dir, epochs, batch_size, weights, single_cls, evolve, data, cfg, resume, noval, nosave, workers, freeze = \
        Path(opt.save_dir), opt.epochs, opt.batch_size, opt.weights, opt.single_cls, opt.evolve, opt.data, opt.cfg, \
        opt.resume, opt.noval, opt.nosave, opt.workers, opt.freeze
    callbacks.run('on_pretrain_routine_start')

    # Directories
    w = save_dir / 'weights'  # weights dir
    (w.parent if evolve else w).mkdir(parents=True, exist_ok=True)  # make dir
    last, best = w / 'last.pt', w / 'best.pt'

    # Hyperparameters
    if isinstance(hyp, str):
        with open(hyp, errors='ignore') as f:
            hyp = yaml.safe_load(f)  # load hyps dict
    LOGGER.info(colorstr('hyperparameters: ') + ', '.join(f'{k}={v}' for k, v in hyp.items()))
    opt.hyp = hyp.copy()  # for saving hyps to checkpoints

    # Save run settings
    if not evolve:
        yaml_save(save_dir / 'hyp.yaml', hyp)
        yaml_save(save_dir / 'opt.yaml', vars(opt))

    # Loggers
    data_dict = None
    if RANK in {-1, 0}:
        loggers = Loggers(save_dir, weights, opt, hyp, LOGGER)  # loggers instance

        # Register actions
        for k in methods(loggers):
            callbacks.register_action(k, callback=getattr(loggers, k))

        # Process custom dataset artifact link
        data_dict = loggers.remote_dataset
        if resume:  # If resuming runs from remote artifact
            weights, epochs, hyp, batch_size = opt.weights, opt.epochs, opt.hyp, opt.batch_size

    # Config
    plots = not evolve and not opt.noplots  # create plots
    cuda = device.type != 'cpu'
    init_seeds(opt.seed + 1 + RANK, deterministic=True)
    with torch_distributed_zero_first(LOCAL_RANK):
        data_dict = data_dict or check_dataset(data)  # check if None
    train_path, val_path = data_dict['train'], data_dict['val']
    nc = 1 if single_cls else int(data_dict['nc'])  # number of classes
    names = {0: 'item'} if single_cls and len(data_dict['names']) != 1 else data_dict['names']  # class names
    is_coco = isinstance(val_path, str) and val_path.endswith('coco/val2017.txt')  # COCO dataset

    # Model
    check_suffix(weights, '.pt')  # check weights
    pretrained = weights.endswith('.pt')
    if pretrained:
        with torch_distributed_zero_first(LOCAL_RANK):
            weights = attempt_download(weights)  # download if not found locally
        ckpt = torch.load(weights, map_location='cpu')  # load checkpoint to CPU to avoid CUDA memory leak
        model = Model(cfg or ckpt['model'].yaml, ch=3, nc=nc, anchors=hyp.get('anchors')).to(device)  # create
        exclude = ['anchor'] if (cfg or hyp.get('anchors')) and not resume else []  # exclude keys
        csd = ckpt['model'].float().state_dict()  # checkpoint state_dict as FP32
        csd = intersect_dicts(csd, model.state_dict(), exclude=exclude)  # intersect
        model.load_state_dict(csd, strict=False)  # load
        LOGGER.info(f'Transferred {len(csd)}/{len(model.state_dict())} items from {weights}')  # report
    else:
        model = Model(cfg, ch=3, nc=nc, anchors=hyp.get('anchors')).to(device)  # create
    amp = check_amp(model)  # check AMP

    # Freeze
    freeze = [f'model.{x}.' for x in (freeze if len(freeze) > 1 else range(freeze[0]))]  # layers to freeze
    for k, v in model.named_parameters():
        v.requires_grad = True  # train all layers
        # v.register_hook(lambda x: torch.nan_to_num(x))  # NaN to 0 (commented for erratic training results)
        if any(x in k for x in freeze):
            LOGGER.info(f'freezing {k}')
            v.requires_grad = False

    # Image size
    gs = max(int(model.stride.max()), 32)  # grid size (max stride)
    imgsz = check_img_size(opt.imgsz, gs, floor=gs * 2)  # verify imgsz is gs-multiple

    # Batch size
    if RANK == -1 and batch_size == -1:  # single-GPU only, estimate best batch size
        batch_size = check_train_batch_size(model, imgsz, amp)
        loggers.on_params_update({"batch_size": batch_size})

    # Optimizer
    nbs = 64  # nominal batch size
    accumulate = max(round(nbs / batch_size), 1)  # accumulate loss before optimizing
    hyp['weight_decay'] *= batch_size * accumulate / nbs  # scale weight_decay
    optimizer = smart_optimizer(model, opt.optimizer, hyp['lr0'], hyp['momentum'], hyp['weight_decay'])

    # Scheduler
    if opt.cos_lr:
        lf = one_cycle(1, hyp['lrf'], epochs)  # cosine 1->hyp['lrf']
    else:
        lf = lambda x: (1 - x / epochs) * (1.0 - hyp['lrf']) + hyp['lrf']  # linear
    scheduler = lr_scheduler.LambdaLR(optimizer, lr_lambda=lf)  # plot_lr_scheduler(optimizer, scheduler, epochs)

    # EMA
    ema = ModelEMA(model) if RANK in {-1, 0} else None

    # Resume
    best_fitness, start_epoch = 0.0, 0
    if pretrained:
        if resume:
            best_fitness, start_epoch, epochs = smart_resume(ckpt, optimizer, ema, weights, epochs, resume)
        del ckpt, csd

    # DP mode
    if cuda and RANK == -1 and torch.cuda.device_count() > 1:
        LOGGER.warning('WARNING ⚠️ DP not recommended, use torch.distributed.run for best DDP Multi-GPU results.\n'
                       'See Multi-GPU Tutorial at https://github.com/ultralytics/yolov5/issues/475 to get started.')
        model = torch.nn.DataParallel(model)

    # SyncBatchNorm
    if opt.sync_bn and cuda and RANK != -1:
        model = torch.nn.SyncBatchNorm.convert_sync_batchnorm(model).to(device)
        LOGGER.info('Using SyncBatchNorm()')

    # Trainloader
    train_loader, dataset = create_dataloader(train_path,
                                              imgsz,
                                              batch_size // WORLD_SIZE,
                                              gs,
                                              single_cls,
                                              hyp=hyp,
                                              augment=True,
                                              cache=None if opt.cache == 'val' else opt.cache,
                                              rect=opt.rect,
                                              rank=LOCAL_RANK,
                                              workers=workers,
                                              image_weights=opt.image_weights,
                                              quad=opt.quad,
                                              prefix=colorstr('train: '),
                                              shuffle=True)
    labels = np.concatenate(dataset.labels, 0)
    mlc = int(labels[:, 0].max())  # max label class
    assert mlc < nc, f'Label class {mlc} exceeds nc={nc} in {data}. Possible class labels are 0-{nc - 1}'

    # Process 0
    if RANK in {-1, 0}:
        val_loader = create_dataloader(val_path,
                                       imgsz,
                                       batch_size // WORLD_SIZE * 2,
                                       gs,
                                       single_cls,
                                       hyp=hyp,
                                       cache=None if noval else opt.cache,
                                       rect=True,
                                       rank=-1,
                                       workers=workers * 2,
                                       pad=0.5,
                                       prefix=colorstr('val: '))[0]

        if not resume:
            if not opt.noautoanchor:
                check_anchors(dataset, model=model, thr=hyp['anchor_t'], imgsz=imgsz)  # run AutoAnchor
            model.half().float()  # pre-reduce anchor precision

        callbacks.run('on_pretrain_routine_end', labels, names)

    # DDP mode
    if cuda and RANK != -1:
        model = smart_DDP(model)

    # Model attributes
    nl = de_parallel(model).model[-1].nl  # number of detection layers (to scale hyps)
    hyp['box'] *= 3 / nl  # scale to layers
    hyp['cls'] *= nc / 80 * 3 / nl  # scale to classes and layers
    hyp['obj'] *= (imgsz / 640) ** 2 * 3 / nl  # scale to image size and layers
    hyp['label_smoothing'] = opt.label_smoothing
    model.nc = nc  # attach number of classes to model
    model.hyp = hyp  # attach hyperparameters to model
    model.class_weights = labels_to_class_weights(dataset.labels, nc).to(device) * nc  # attach class weights
    model.names = names

    # Start training
    t0 = time.time()
    nb = len(train_loader)  # number of batches
    nw = max(round(hyp['warmup_epochs'] * nb), 100)  # number of warmup iterations, max(3 epochs, 100 iterations)
    # nw = min(nw, (epochs - start_epoch) / 2 * nb)  # limit warmup to < 1/2 of training
    last_opt_step = -1
    maps = np.zeros(nc)  # mAP per class
    results = (0, 0, 0, 0, 0, 0, 0)  # P, R, mAP@.5, mAP@.5-.95, val_loss(box, obj, cls)
    scheduler.last_epoch = start_epoch - 1  # do not move
    scaler = torch.cuda.amp.GradScaler(enabled=amp)
    stopper, stop = EarlyStopping(patience=opt.patience), False
    compute_loss = ComputeLoss(model)  # init loss class
    callbacks.run('on_train_start')
    LOGGER.info(f'Image sizes {imgsz} train, {imgsz} val\n'
                f'Using {train_loader.num_workers * WORLD_SIZE} dataloader workers\n'
                f"Logging results to {colorstr('bold', save_dir)}\n"
                f'Starting training for {epochs} epochs...')
    for epoch in range(start_epoch, epochs):  # epoch ------------------------------------------------------------------
        callbacks.run('on_train_epoch_start')
        model.train()

        # Update image weights (optional, single-GPU only)
        if opt.image_weights:
            cw = model.class_weights.cpu().numpy() * (1 - maps) ** 2 / nc  # class weights
            iw = labels_to_image_weights(dataset.labels, nc=nc, class_weights=cw)  # image weights
            dataset.indices = random.choices(range(dataset.n), weights=iw, k=dataset.n)  # rand weighted idx

        # Update mosaic border (optional)
        # b = int(random.uniform(0.25 * imgsz, 0.75 * imgsz + gs) // gs * gs)
        # dataset.mosaic_border = [b - imgsz, -b]  # height, width borders

        mloss = torch.zeros(3, device=device)  # mean losses
        if RANK != -1:
            train_loader.sampler.set_epoch(epoch)
        pbar = enumerate(train_loader)
        LOGGER.info(('\n' + '%11s' * 7) % ('Epoch', 'GPU_mem', 'box_loss', 'obj_loss', 'cls_loss', 'Instances', 'Size'))
        if RANK in {-1, 0}:
            pbar = tqdm(pbar, total=nb, bar_format=TQDM_BAR_FORMAT)  # progress bar
        optimizer.zero_grad()
        for i, (imgs, targets, paths, _) in pbar:  # batch -------------------------------------------------------------
            callbacks.run('on_train_batch_start')
            ni = i + nb * epoch  # number integrated batches (since train start)
            imgs = imgs.to(device, non_blocking=True).float() / 255  # uint8 to float32, 0-255 to 0.0-1.0

            # Warmup
            if ni <= nw:
                xi = [0, nw]  # x interp
                # compute_loss.gr = np.interp(ni, xi, [0.0, 1.0])  # iou loss ratio (obj_loss = 1.0 or iou)
                accumulate = max(1, np.interp(ni, xi, [1, nbs / batch_size]).round())
                for j, x in enumerate(optimizer.param_groups):
                    # bias lr falls from 0.1 to lr0, all other lrs rise from 0.0 to lr0
                    x['lr'] = np.interp(ni, xi, [hyp['warmup_bias_lr'] if j == 0 else 0.0, x['initial_lr'] * lf(epoch)])
                    if 'momentum' in x:
                        x['momentum'] = np.interp(ni, xi, [hyp['warmup_momentum'], hyp['momentum']])

            # Multi-scale
            if opt.multi_scale:
                sz = random.randrange(imgsz * 0.5, imgsz * 1.5 + gs) // gs * gs  # size
                sf = sz / max(imgs.shape[2:])  # scale factor
                if sf != 1:
                    ns = [math.ceil(x * sf / gs) * gs for x in imgs.shape[2:]]  # new shape (stretched to gs-multiple)
                    imgs = nn.functional.interpolate(imgs, size=ns, mode='bilinear', align_corners=False)

            # Forward
            with torch.cuda.amp.autocast(amp):
                pred = model(imgs)  # forward
                loss, loss_items = compute_loss(pred, targets.to(device))  # loss scaled by batch_size
                if RANK != -1:
                    loss *= WORLD_SIZE  # gradient averaged between devices in DDP mode
                if opt.quad:
                    loss *= 4.

            # Backward
            scaler.scale(loss).backward()

            # Optimize - https://pytorch.org/docs/master/notes/amp_examples.html
            if ni - last_opt_step >= accumulate:
                scaler.unscale_(optimizer)  # unscale gradients
                torch.nn.utils.clip_grad_norm_(model.parameters(), max_norm=10.0)  # clip gradients
                scaler.step(optimizer)  # optimizer.step
                scaler.update()
                optimizer.zero_grad()
                if ema:
                    ema.update(model)
                last_opt_step = ni

            # Log
            if RANK in {-1, 0}:
                mloss = (mloss * i + loss_items) / (i + 1)  # update mean losses
                mem = f'{torch.cuda.memory_reserved() / 1E9 if torch.cuda.is_available() else 0:.3g}G'  # (GB)
                pbar.set_description(('%11s' * 2 + '%11.4g' * 5) %
                                     (f'{epoch}/{epochs - 1}', mem, *mloss, targets.shape[0], imgs.shape[-1]))
                callbacks.run('on_train_batch_end', model, ni, imgs, targets, paths, list(mloss))
                if callbacks.stop_training:
                    return
            # end batch ------------------------------------------------------------------------------------------------

        # Scheduler
        lr = [x['lr'] for x in optimizer.param_groups]  # for loggers
        scheduler.step()

        if RANK in {-1, 0}:
            # mAP
            callbacks.run('on_train_epoch_end', epoch=epoch)
            ema.update_attr(model, include=['yaml', 'nc', 'hyp', 'names', 'stride', 'class_weights'])
            final_epoch = (epoch + 1 == epochs) or stopper.possible_stop
            if not noval or final_epoch:  # Calculate mAP
                results, maps, _ = validate.run(data_dict,
                                                batch_size=batch_size // WORLD_SIZE * 2,
                                                imgsz=imgsz,
                                                half=amp,
                                                model=ema.ema,
                                                single_cls=single_cls,
                                                dataloader=val_loader,
                                                save_dir=save_dir,
                                                plots=False,
                                                callbacks=callbacks,
                                                compute_loss=compute_loss)

            # Update best mAP
            fi = fitness(np.array(results).reshape(1, -1))  # weighted combination of [P, R, mAP@.5, mAP@.5-.95]
            stop = stopper(epoch=epoch, fitness=fi)  # early stop check
            if fi > best_fitness:
                best_fitness = fi
            log_vals = list(mloss) + list(results) + lr
            callbacks.run('on_fit_epoch_end', log_vals, epoch, best_fitness, fi)

            # Save model
            if (not nosave) or (final_epoch and not evolve):  # if save
                ckpt = {
                    'epoch': epoch,
                    'best_fitness': best_fitness,
                    'model': deepcopy(de_parallel(model)).half(),
                    'ema': deepcopy(ema.ema).half(),
                    'updates': ema.updates,
                    'optimizer': optimizer.state_dict(),
                    'opt': vars(opt),
                    'git': GIT_INFO,  # {remote, branch, commit} if a git repo
                    'date': datetime.now().isoformat()}

                # Save last, best and delete
                torch.save(ckpt, last)
                if best_fitness == fi:
                    torch.save(ckpt, best)
                if opt.save_period > 0 and epoch % opt.save_period == 0:
                    torch.save(ckpt, w / f'epoch{epoch}.pt')
                del ckpt
                callbacks.run('on_model_save', last, epoch, final_epoch, best_fitness, fi)

        # EarlyStopping
        if RANK != -1:  # if DDP training
            broadcast_list = [stop if RANK == 0 else None]
            dist.broadcast_object_list(broadcast_list, 0)  # broadcast 'stop' to all ranks
            if RANK != 0:
                stop = broadcast_list[0]
        if stop:
            break  # must break all DDP ranks

        # end epoch ----------------------------------------------------------------------------------------------------
    # end training -----------------------------------------------------------------------------------------------------
    if RANK in {-1, 0}:
        LOGGER.info(f'\n{epoch - start_epoch + 1} epochs completed in {(time.time() - t0) / 3600:.3f} hours.')
        for f in last, best:
            if f.exists():
                strip_optimizer(f)  # strip optimizers
                if f is best:
                    LOGGER.info(f'\nValidating {f}...')
                    results, _, _ = validate.run(
                        data_dict,
                        batch_size=batch_size // WORLD_SIZE * 2,
                        imgsz=imgsz,
                        model=attempt_load(f, device).half(),
                        iou_thres=0.65 if is_coco else 0.60,  # best pycocotools at iou 0.65
                        single_cls=single_cls,
                        dataloader=val_loader,
                        save_dir=save_dir,
                        save_json=is_coco,
                        verbose=True,
                        plots=plots,
                        callbacks=callbacks,
                        compute_loss=compute_loss)  # val best model with plots
                    if is_coco:
                        callbacks.run('on_fit_epoch_end', list(mloss) + list(results) + lr, epoch, best_fitness, fi)

        callbacks.run('on_train_end', last, best, epoch, results)

    torch.cuda.empty_cache()
    return results


def parse_opt(known=False):
    parser = argparse.ArgumentParser()
    parser.add_argument('--weights', type=str, default=ROOT / 'yolov5s.pt', help='初始 weights 路徑')
    parser.add_argument('--cfg', type=str, default='', help='model.yaml 路徑')
    parser.add_argument('--data', type=str, default=ROOT / 'data/coco128.yaml', help='dataset.yaml 路徑')
    parser.add_argument('--hyp', type=str, default=ROOT / 'data/hyps/hyp.scratch-low.yaml', help='hyperparameters 路徑')
    parser.add_argument('--epochs', type=int, default=100, help='總訓練時期')
    parser.add_argument('--batch-size', type=int, default=16, help='所有 GPU 的總批次大小，-1 表示自動批次')
    parser.add_argument('--imgsz', '--img', '--img-size', type=int, default=640, help='train, val 圖像大小（像素）')
    parser.add_argument('--rect', action='store_true', help='矩形訓練')
    parser.add_argument('--resume', nargs='?', const=True, default=False, help='恢復最近的訓練')
    parser.add_argument('--nosave', action='store_true', help='只保存最後的檢查點')
    parser.add_argument('--noval', action='store_true', help='只驗證最終epoch')
    parser.add_argument('--noautoanchor', action='store_true', help='禁用自動錨點')
    parser.add_argument('--noplots', action='store_true', help='不保存繪圖文件')
    parser.add_argument('--evolve', type=int, nargs='?', const=300, help='evolve hyperparameters for x generations')
    parser.add_argument('--bucket', type=str, default='', help='gsutil bucket')
    parser.add_argument('--cache', type=str, nargs='?', const='ram', help='圖像--緩存內存/磁盤')
    parser.add_argument('--image-weights', action='store_true', help='使用weighted圖像選擇進行訓練')
    parser.add_argument('--device', default='', help='cuda 設備，即 0 或 0,1,2,3 或 cpu')
    parser.add_argument('--multi-scale', action='store_true', help='vary img-size +/- 50%%')
    parser.add_argument('--single-cls', action='store_true', help='將多類數據訓練為單類')
    parser.add_argument('--optimizer', type=str, choices=['SGD', 'Adam', 'AdamW'], default='SGD', help='優化器')
    parser.add_argument('--sync-bn', action='store_true', help='使用 SyncBatchNorm，僅在 DDP 模式下可用')
    parser.add_argument('--workers', type=int, default=8, help='最大數據加載器工作人員（在 DDP 模式下按 RANK）')
    parser.add_argument('--project', default=ROOT / 'runs/train', help='保存到項目/名稱')
    parser.add_argument('--name', default='exp', help='保存到項目/名稱')
    parser.add_argument('--exist-ok', action='store_true', help='現有項目/名稱OK，不要增加')
    parser.add_argument('--quad', action='store_true', help='quad dataloader')
    parser.add_argument('--cos-lr', action='store_true', help='cosine LR scheduler')
    parser.add_argument('--label-smoothing', type=float, default=0.0, help='Label smoothing epsilon')
    parser.add_argument('--patience', type=int, default=100, help='EarlyStopping patience (epochs without improvement)')
    parser.add_argument('--freeze', nargs='+', type=int, default=[0], help='Freeze layers: backbone=10, first3=0 1 2')
    parser.add_argument('--save-period', type=int, default=-1, help='Save checkpoint every x epochs (disabled if < 1)')
    parser.add_argument('--seed', type=int, default=0, help='Global training seed')
    parser.add_argument('--local_rank', type=int, default=-1, help='自動 DDP Multi-GPU 參數，請勿修改')

    # 記錄器參數
    parser.add_argument('--entity', default=None, help='實體')
    parser.add_argument('--upload_dataset', nargs='?', const=True, default=False, help='上傳數據，“val”選項')
    parser.add_argument('--bbox_interval', type=int, default=-1, help='設置邊界框圖像記錄間隔')
    parser.add_argument('--artifact_alias', type=str, default='latest', help='要使用的數據集工件版本')

    return parser.parse_known_args()[0] if known else parser.parse_args()


def main(opt, callbacks=Callbacks()):
    # Checks
    if RANK in {-1, 0}:
        print_args(vars(opt))
        check_git_status()
        check_requirements()

    # Resume (from specified or most recent last.pt)
    if opt.resume and not check_comet_resume(opt) and not opt.evolve:
        last = Path(check_file(opt.resume) if isinstance(opt.resume, str) else get_latest_run())
        opt_yaml = last.parent.parent / 'opt.yaml'  # train options yaml
        opt_data = opt.data  # original dataset
        if opt_yaml.is_file():
            with open(opt_yaml, errors='ignore') as f:
                d = yaml.safe_load(f)
        else:
            d = torch.load(last, map_location='cpu')['opt']
        opt = argparse.Namespace(**d)  # replace
        opt.cfg, opt.weights, opt.resume = '', str(last), True  # reinstate
        if is_url(opt_data):
            opt.data = check_file(opt_data)  # avoid HUB resume auth timeout
    else:
        opt.data, opt.cfg, opt.hyp, opt.weights, opt.project = \
            check_file(opt.data), check_yaml(opt.cfg), check_yaml(opt.hyp), str(opt.weights), str(opt.project)  # checks
        assert len(opt.cfg) or len(opt.weights), 'either --cfg or --weights must be specified'
        if opt.evolve:
            if opt.project == str(ROOT / 'runs/train'):  # if default project name, rename to runs/evolve
                opt.project = str(ROOT / 'runs/evolve')
            opt.exist_ok, opt.resume = opt.resume, False  # pass resume to exist_ok and disable resume
        if opt.name == 'cfg':
            opt.name = Path(opt.cfg).stem  # use model.yaml as name
        opt.save_dir = str(increment_path(Path(opt.project) / opt.name, exist_ok=opt.exist_ok))

    # DDP mode
    device = select_device(opt.device, batch_size=opt.batch_size)
    if LOCAL_RANK != -1:
        msg = 'is not compatible with YOLOv5 Multi-GPU DDP training'
        assert not opt.image_weights, f'--image-weights {msg}'
        assert not opt.evolve, f'--evolve {msg}'
        assert opt.batch_size != -1, f'AutoBatch with --batch-size -1 {msg}, please pass a valid --batch-size'
        assert opt.batch_size % WORLD_SIZE == 0, f'--batch-size {opt.batch_size} must be multiple of WORLD_SIZE'
        assert torch.cuda.device_count() > LOCAL_RANK, 'insufficient CUDA devices for DDP command'
        torch.cuda.set_device(LOCAL_RANK)
        device = torch.device('cuda', LOCAL_RANK)
        dist.init_process_group(backend="nccl" if dist.is_nccl_available() else "gloo")

    # Train
    if not opt.evolve:
        train(opt.hyp, opt, device, callbacks)

    # Evolve hyperparameters (optional)
    else:
        # Hyperparameter evolution metadata (mutation scale 0-1, lower_limit, upper_limit)
        meta = {
            'lr0': (1, 1e-5, 1e-1),  # initial learning rate (SGD=1E-2, Adam=1E-3)
            'lrf': (1, 0.01, 1.0),  # final OneCycleLR learning rate (lr0 * lrf)
            'momentum': (0.3, 0.6, 0.98),  # SGD momentum/Adam beta1
            'weight_decay': (1, 0.0, 0.001),  # optimizer weight decay
            'warmup_epochs': (1, 0.0, 5.0),  # warmup epochs (fractions ok)
            'warmup_momentum': (1, 0.0, 0.95),  # warmup initial momentum
            'warmup_bias_lr': (1, 0.0, 0.2),  # warmup initial bias lr
            'box': (1, 0.02, 0.2),  # box loss gain
            'cls': (1, 0.2, 4.0),  # cls loss gain
            'cls_pw': (1, 0.5, 2.0),  # cls BCELoss positive_weight
            'obj': (1, 0.2, 4.0),  # obj loss gain (scale with pixels)
            'obj_pw': (1, 0.5, 2.0),  # obj BCELoss positive_weight
            'iou_t': (0, 0.1, 0.7),  # IoU training threshold
            'anchor_t': (1, 2.0, 8.0),  # anchor-multiple threshold
            'anchors': (2, 2.0, 10.0),  # anchors per output grid (0 to ignore)
            'fl_gamma': (0, 0.0, 2.0),  # focal loss gamma (efficientDet default gamma=1.5)
            'hsv_h': (1, 0.0, 0.1),  # image HSV-Hue augmentation (fraction)
            'hsv_s': (1, 0.0, 0.9),  # image HSV-Saturation augmentation (fraction)
            'hsv_v': (1, 0.0, 0.9),  # image HSV-Value augmentation (fraction)
            'degrees': (1, 0.0, 45.0),  # image rotation (+/- deg)
            'translate': (1, 0.0, 0.9),  # image translation (+/- fraction)
            'scale': (1, 0.0, 0.9),  # image scale (+/- gain)
            'shear': (1, 0.0, 10.0),  # image shear (+/- deg)
            'perspective': (0, 0.0, 0.001),  # image perspective (+/- fraction), range 0-0.001
            'flipud': (1, 0.0, 1.0),  # image flip up-down (probability)
            'fliplr': (0, 0.0, 1.0),  # image flip left-right (probability)
            'mosaic': (1, 0.0, 1.0),  # image mixup (probability)
            'mixup': (1, 0.0, 1.0),  # image mixup (probability)
            'copy_paste': (1, 0.0, 1.0)}  # segment copy-paste (probability)

        with open(opt.hyp, errors='ignore') as f:
            hyp = yaml.safe_load(f)  # load hyps dict
            if 'anchors' not in hyp:  # anchors commented in hyp.yaml
                hyp['anchors'] = 3
        if opt.noautoanchor:
            del hyp['anchors'], meta['anchors']
        opt.noval, opt.nosave, save_dir = True, True, Path(opt.save_dir)  # only val/save final epoch
        # ei = [isinstance(x, (int, float)) for x in hyp.values()]  # evolvable indices
        evolve_yaml, evolve_csv = save_dir / 'hyp_evolve.yaml', save_dir / 'evolve.csv'
        if opt.bucket:
            os.system(f'gsutil cp gs://{opt.bucket}/evolve.csv {evolve_csv}')  # download evolve.csv if exists

        for _ in range(opt.evolve):  # generations to evolve
            if evolve_csv.exists():  # if evolve.csv exists: select best hyps and mutate
                # Select parent(s)
                parent = 'single'  # parent selection method: 'single' or 'weighted'
                x = np.loadtxt(evolve_csv, ndmin=2, delimiter=',', skiprows=1)
                n = min(5, len(x))  # number of previous results to consider
                x = x[np.argsort(-fitness(x))][:n]  # top n mutations
                w = fitness(x) - fitness(x).min() + 1E-6  # weights (sum > 0)
                if parent == 'single' or len(x) == 1:
                    # x = x[random.randint(0, n - 1)]  # random selection
                    x = x[random.choices(range(n), weights=w)[0]]  # weighted selection
                elif parent == 'weighted':
                    x = (x * w.reshape(n, 1)).sum(0) / w.sum()  # weighted combination

                # Mutate
                mp, s = 0.8, 0.2  # mutation probability, sigma
                npr = np.random
                npr.seed(int(time.time()))
                g = np.array([meta[k][0] for k in hyp.keys()])  # gains 0-1
                ng = len(meta)
                v = np.ones(ng)
                while all(v == 1):  # mutate until a change occurs (prevent duplicates)
                    v = (g * (npr.random(ng) < mp) * npr.randn(ng) * npr.random() * s + 1).clip(0.3, 3.0)
                for i, k in enumerate(hyp.keys()):  # plt.hist(v.ravel(), 300)
                    hyp[k] = float(x[i + 7] * v[i])  # mutate

            # Constrain to limits
            for k, v in meta.items():
                hyp[k] = max(hyp[k], v[1])  # lower limit
                hyp[k] = min(hyp[k], v[2])  # upper limit
                hyp[k] = round(hyp[k], 5)  # significant digits

            # Train mutation
            results = train(hyp.copy(), opt, device, callbacks)
            callbacks = Callbacks()
            # Write mutation results
            keys = ('metrics/precision', 'metrics/recall', 'metrics/mAP_0.5', 'metrics/mAP_0.5:0.95', 'val/box_loss',
                    'val/obj_loss', 'val/cls_loss')
            print_mutation(keys, results, hyp.copy(), save_dir, opt.bucket)

        # Plot results
        plot_evolve(evolve_csv)
        LOGGER.info(f'Hyperparameter evolution finished {opt.evolve} generations\n'
                    f"Results saved to {colorstr('bold', save_dir)}\n"
                    f'Usage example: $ python train.py --hyp {evolve_yaml}')


def run(**kwargs):
    # Usage: import train; train.run(data='coco128.yaml', imgsz=320, weights='yolov5m.pt')
    opt = parse_opt(True)
    for k, v in kwargs.items():
        setattr(opt, k, v)
    main(opt)
    return opt


if __name__ == "__main__":
    opt = parse_opt()
    main(opt)

# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
"""
運行YOLOv5檢測推斷圖像，視頻，目錄，glob, YouTube，網絡攝像頭，etc等。

用法 - 來源：
    $ python detect.py --weights yolov5s.pt --source 0                               # webcam
                                                     img.jpg                         # image
                                                     vid.mp4                         # video
                                                     screen                          # screenshot
                                                     path/                           # directory
                                                     'path/*.jpg'                    # glob
                                                     'https://youtu.be/Zgi9g1ksQHc'  # YouTube
                                                     'rtsp://example.com/media.mp4'  # RTSP, RTMP, HTTP stream

用法 - 格式：
    $ python detect.py --weights yolov5s.pt                 # PyTorch
                                 yolov5s.torchscript        # TorchScript
                                 yolov5s.onnx               # ONNX Runtime or OpenCV DNN with --dnn
                                 yolov5s_openvino_model     # OpenVINO
                                 yolov5s.engine             # TensorRT
                                 yolov5s.mlmodel            # CoreML (macOS-only)
                                 yolov5s_saved_model        # TensorFlow SavedModel
                                 yolov5s.pb                 # TensorFlow GraphDef
                                 yolov5s.tflite             # TensorFlow Lite
                                 yolov5s_edgetpu.tflite     # TensorFlow Edge TPU
                                 yolov5s_paddle_model       # PaddlePaddle
"""

import argparse
import os
import platform
import sys
from pathlib import Path

import torch

FILE = Path(__file__).resolve()
ROOT = FILE.parents[0]  # YOLOv5 root directory
if str(ROOT) not in sys.path:
    sys.path.append(str(ROOT))  # 將 ROOT 添加到 PATH
ROOT = Path(os.path.relpath(ROOT, Path.cwd()))  # relative

from models.common import DetectMultiBackend
from utils.dataloaders import IMG_FORMATS, VID_FORMATS, LoadImages, LoadScreenshots, LoadStreams
from utils.general import (LOGGER, Profile, check_file, check_img_size, check_imshow, check_requirements, colorstr, cv2,
                           increment_path, non_max_suppression, print_args, scale_boxes, strip_optimizer, xyxy2xywh)
from utils.plots import Annotator, colors, save_one_box
from utils.torch_utils import select_device, smart_inference_mode


@smart_inference_mode()
def run(
        weights=ROOT / 'yolov5s.pt',  # 模型路徑或 triton URL
        source=ROOT / 'data/images',  # 文件/目錄/URL/glob/screen/0（網絡攝像頭）
        data=ROOT / 'data/coco128.yaml',  # dataset.yaml路徑
        imgsz=(640, 640),  # 推理大小（高度, 寬度）
        conf_thres=0.25,  # confidence 閾值
        iou_thres=0.45,  # NMS IOU 閾值
        max_det=1000,  # 每張圖片的最大檢測數
        device='',  # cuda 設備，即 0 或 0,1,2,3 或 cpu
        view_img=False,  # 顯示結果
        save_txt=False,  # 將結果保存到 *.txt
        save_conf=False,  # 在 --save-txt 標籤中保存信心
        save_crop=False,  # 保存裁剪的預測框
        nosave=False,  # 不保存圖片/視頻
        classes=None,  # 按類別過濾：--class 0，或--class 0 2 3
        agnostic_nms=False,  # class-agnostic NMS
        augment=False,  # 增強推理
        visualize=False,  # 可視化特徵
        update=False,  # 更新所有模型
        project=ROOT / 'runs/detect',  # 將結果保存到項目/名稱
        name='exp',  # 將結果保存到項目/名稱
        exist_ok=False,  # 現有項目/名稱可以，不要增加
        line_thickness=3,  # 邊界框厚度（pixels像素）
        hide_labels=False,  # 隱藏標籤
        hide_conf=False,  # hide confidences
        half=False,  # 使用 FP16 半精度推理
        dnn=False,  # 使用 OpenCV DNN 進行 ONNX 推理
        vid_stride=1,  # 視頻幀率步幅
):
    source = str(source)
    save_img = not nosave and not source.endswith('.txt')  # 保存推理圖像
    is_file = Path(source).suffix[1:] in (IMG_FORMATS + VID_FORMATS)
    is_url = source.lower().startswith(('rtsp://', 'rtmp://', 'http://', 'https://'))
    webcam = source.isnumeric() or source.endswith('.txt') or (is_url and not is_file)
    screenshot = source.lower().startswith('screen')
    if is_url and is_file:
        source = check_file(source)  # download

    # 目錄
    save_dir = increment_path(Path(project) / name, exist_ok=exist_ok)  # increment run
    (save_dir / 'labels' if save_txt else save_dir).mkdir(parents=True, exist_ok=True)  # make dir

    # 加載模型
    device = select_device(device)
    model = DetectMultiBackend(weights, device=device, dnn=dnn, data=data, fp16=half)
    stride, names, pt = model.stride, model.names, model.pt
    imgsz = check_img_size(imgsz, s=stride)  # 檢查圖像大小

    # 數據加載器
    bs = 1  # batch_size
    if webcam:
        view_img = check_imshow(warn=True)
        dataset = LoadStreams(source, img_size=imgsz, stride=stride, auto=pt, vid_stride=vid_stride)
        bs = len(dataset)
    elif screenshot:
        dataset = LoadScreenshots(source, img_size=imgsz, stride=stride, auto=pt)
    else:
        dataset = LoadImages(source, img_size=imgsz, stride=stride, auto=pt, vid_stride=vid_stride)
    vid_path, vid_writer = [None] * bs, [None] * bs

    # 運行推理
    model.warmup(imgsz=(1 if pt or model.triton else bs, 3, *imgsz))  # warmup
    seen, windows, dt = 0, [], (Profile(), Profile(), Profile())
    for path, im, im0s, vid_cap, s in dataset:
        with dt[0]:
            im = torch.from_numpy(im).to(model.device)
            im = im.half() if model.fp16 else im.float()  # uint8 to fp16/32
            im /= 255  # 0 - 255 to 0.0 - 1.0
            if len(im.shape) == 3:
                im = im[None]  # expand for batch dim

        # 推理
        with dt[1]:
            visualize = increment_path(save_dir / Path(path).stem, mkdir=True) if visualize else False
            pred = model(im, augment=augment, visualize=visualize)

        # NMS
        with dt[2]:
            pred = non_max_suppression(pred, conf_thres, iou_thres, classes, agnostic_nms, max_det=max_det)

        # 第二階段分類器（可選）
        # pred = utils.general.apply_classifier(pred, classifier_model, im, im0s)

        # 過程預測
        for i, det in enumerate(pred):  # 每張圖片
            seen += 1
            if webcam:  # batch_size >= 1
                p, im0, frame = path[i], im0s[i].copy(), dataset.count
                s += f'{i}: '
            else:
                p, im0, frame = path, im0s.copy(), getattr(dataset, 'frame', 0)

            p = Path(p)  # to Path
            save_path = str(save_dir / p.name)  # im.jpg
            txt_path = str(save_dir / 'labels' / p.stem) + ('' if dataset.mode == 'image' else f'_{frame}')  # im.txt
            s += '%gx%g ' % im.shape[2:]  # print string
            gn = torch.tensor(im0.shape)[[1, 0, 1, 0]]  # normalization gain whwh
            imc = im0.copy() if save_crop else im0  # for save_crop
            annotator = Annotator(im0, line_width=line_thickness, example=str(names))
            if len(det):
                # 將框從 img_size 重新縮放為 im0 大小
                det[:, :4] = scale_boxes(im.shape[2:], det[:, :4], im0.shape).round()

                # Print結果
                for c in det[:, 5].unique():
                    n = (det[:, 5] == c).sum()  # 每class檢測
                    s += f"{n} {names[int(c)]}{'s' * (n > 1)}, "  # add to string

                # Write結果
                for *xyxy, conf, cls in reversed(det):

                    if save_txt:  # Write to file
                        xywh = (xyxy2xywh(torch.tensor(xyxy).view(1, 4)) / gn).view(-1).tolist()  # normalized xywh # 歸一化的xywh
                        line = (cls, *xywh, conf) if save_conf else (cls, *xywh)  # label 格式
                        with open(f'{txt_path}.txt', 'a') as f:
                            f.write(('%g ' * len(line)).rstrip() % line + '\n')

                    if save_img or save_crop or view_img:  # Add bbox to image
                        c = int(cls)  # integer class
                        label = None if hide_labels else (names[c] if hide_conf else f'{names[c]} {conf:.2f}')
                        annotator.box_label(xyxy, label, color=colors(c, True))
                    if save_crop:
                        save_one_box(xyxy, imc, file=save_dir / 'crops' / names[c] / f'{p.stem}.jpg', BGR=True)

            # 串流結果
            im0 = annotator.result()
            if view_img:
                if platform.system() == 'Linux' and p not in windows:
                    windows.append(p)
                    cv2.namedWindow(str(p), cv2.WINDOW_NORMAL | cv2.WINDOW_KEEPRATIO)  # 允許調整窗口大小 (Linux)
                    cv2.resizeWindow(str(p), im0.shape[1], im0.shape[0])
                cv2.imshow(str(p), im0) #(視窗名稱,顯示圖片)
                cv2.waitKey(1)  # 1 毫秒.

            # 保存結果（帶檢測的圖像）
            if save_img:
                if dataset.mode == 'image':
                    cv2.imwrite(save_path, im0)
                else:  # 'video' or 'stream'
                    if vid_path[i] != save_path:  # new video
                        vid_path[i] = save_path
                        if isinstance(vid_writer[i], cv2.VideoWriter):
                            vid_writer[i].release()  # release previous video writer
                        if vid_cap:  # video
                            fps = vid_cap.get(cv2.CAP_PROP_FPS)
                            w = int(vid_cap.get(cv2.CAP_PROP_FRAME_WIDTH))
                            h = int(vid_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
                        else:  # stream
                            fps, w, h = 30, im0.shape[1], im0.shape[0]
                        save_path = str(Path(save_path).with_suffix('.mp4'))  # 在結果視頻上強制 *.mp4 後綴
                        vid_writer[i] = cv2.VideoWriter(save_path, cv2.VideoWriter_fourcc(*'mp4v'), fps, (w, h))
                    vid_writer[i].write(im0)

        # 提升在訓練時的效率，生成的"images"和"labels"可以用來加強訓練。
        # 條件是辨識新的圖片或影片，做為新的訓練樣本
        if len(det):
            ok_file="images" #【有】辨識到的圖，存放的資料夾名稱
            img_path = str(save_dir / ok_file / p.stem) +("_"+str(seen)+".jpg")  # im.txt
            if not os.path.isdir(str(save_dir / ok_file)):
                os.makedirs(str(save_dir / ok_file))
            cv2.imwrite(img_path,im0s) #(寫入路徑,圖片) #im0s為辨識前原始圖 #im0為辨識過後有眶圖
            box_file="images_box" #【有】辨識到的圖，存放的資料夾名稱
            img_path = str(save_dir / box_file / p.stem) +("_"+str(seen)+".jpg")  # im.txt
            if not os.path.isdir(str(save_dir / box_file)):
                os.makedirs(str(save_dir / box_file))
            cv2.imwrite(img_path,im0) #(寫入路徑,圖片) #im0s為辨識前原始圖 #im0為辨識過後有眶圖
        else:
            # 可以手動將沒有辨識到的圖，另外進行labelimg處理。
            no_file="no_images" #【沒有】辨識到的圖，存放的資料夾名稱
            img_path = str(save_dir / no_file / p.stem) +("_"+str(seen)+".jpg")  # im.txt
            if not os.path.isdir(str(save_dir / no_file)):
                os.makedirs(str(save_dir / no_file))
            cv2.imwrite(img_path,im0s) #(寫入路徑,圖片) #im0s為辨識前原始圖 #im0為辨識過後有眶圖

        # Print時間（僅推理）
        LOGGER.info(f"{s}{'' if len(det) else '(No detections), '}{dt[1].dt * 1E3:.1f}ms")

    # Print結果
    t = tuple(x.t / seen * 1E3 for x in dt)  # speeds per image # 每幅圖像的速度
    LOGGER.info(f'Speed: %.1fms pre-process, %.1fms inference, %.1fms NMS per image at shape {(1, 3, *imgsz)}' % t)
    if save_txt or save_img:
        s = f"\n{len(list(save_dir.glob('labels/*.txt')))} labels saved to {save_dir / 'labels'}" if save_txt else ''
        LOGGER.info(f"Results saved to {colorstr('bold', save_dir)}{s}")
    if update:
        strip_optimizer(weights[0])  # 更新模型（修復 SourceChangeWarning）


def parse_opt():
    parser = argparse.ArgumentParser()
    parser.add_argument('--weights', nargs='+', type=str, default=ROOT / 'yolov5s.pt', help='（可選）模型路徑或 triton URL')
    parser.add_argument('--source', type=str, default=ROOT / 'data/images', help='文件/目錄/URL/glob/screen/0（網絡攝像頭）')
    parser.add_argument('--data', type=str, default=ROOT / 'data/coco128.yaml', help='（可選）dataset.yaml 路徑')
    parser.add_argument('--imgsz', '--img', '--img-size', nargs='+', type=int, default=[480,640], help='推理大小 h,w')
    parser.add_argument('--conf-thres', type=float, default=0.25, help='confidence 閾值')
    parser.add_argument('--iou-thres', type=float, default=0.45, help='NMS IoU 閾值')
    parser.add_argument('--max-det', type=int, default=1000, help='每張圖片的最大檢測數')
    parser.add_argument('--device', default='', help='cuda 設備，即 0 或 0,1,2,3 或 cpu')
    parser.add_argument('--view-img', action='store_true', help='顯示辨識結果')
    parser.add_argument('--save-txt', action='store_true', help='將結果保存到 *.txt')
    parser.add_argument('--save-conf', action='store_true', help='在 --save-txt 標籤中保存信心')
    parser.add_argument('--save-crop', action='store_true', help='保存裁剪的預測框')
    parser.add_argument('--nosave', action='store_true', help='不保存圖片/視頻')
    parser.add_argument('--classes', nargs='+', type=int, help='按類別過濾：--classes 0，或--classes 0 2 3')
    parser.add_argument('--agnostic-nms', action='store_true', help='class-agnostic NMS')
    parser.add_argument('--augment', action='store_true', help='增強推理')
    parser.add_argument('--visualize', action='store_true', help='可視化特徵')
    parser.add_argument('--update', action='store_true', help='更新所有模型')
    parser.add_argument('--project', default=ROOT / 'runs/detect', help='將結果保存到項目/名稱')
    parser.add_argument('--name', default='exp', help='將結果保存到項目/名稱')
    parser.add_argument('--exist-ok', action='store_true', help='現有項目/名稱可以，不要增加')
    parser.add_argument('--line-thickness', default=1, type=int, help='邊界框厚度（pixels像素）')
    parser.add_argument('--hide-labels', default=False, action='store_true', help='隱藏標籤')
    parser.add_argument('--hide-conf', default=False, action='store_true', help='隱藏confidences')
    parser.add_argument('--half', action='store_true', help='使用FP16半精度推斷')
    parser.add_argument('--dnn', action='store_true', help='使用OpenCV DNN進行ONNX推斷')
    parser.add_argument('--vid-stride', type=int, default=1, help='視頻幀率步幅')
    opt = parser.parse_args()
    opt.imgsz *= 2 if len(opt.imgsz) == 1 else 1  # expand
    print_args(vars(opt))
    return opt


def main(opt):
    check_requirements(exclude=('tensorboard', 'thop'))
    run(**vars(opt))


if __name__ == "__main__":
    opt = parse_opt()
    main(opt)