def train(hyp, opt, device, tb_writer=None):
logger.info(colorstr('hyperparameters: ') + ', '.join(f'{k}={v}' for k, v in hyp.items())) # 打印超参数
save_dir, epochs, batch_size, total_batch_size, weights, rank = \
Path(opt.save_dir), opt.epochs, opt.batch_size, opt.total_batch_size, opt.weights, opt.global_rank
# Directories
wdir = save_dir / 'weights' # runs/train/exp_/weights
wdir.mkdir(parents=True, exist_ok=True) # make dir,判断该路径是否存在，如果不存在则创建此路径
last = wdir / 'last.pt' # runs/train/exp23/weights/last.pt
best = wdir / 'best.pt' # runs/train/exp23/weights/best.pt
results_file = save_dir / 'results.txt' # runs/train/exp23/results.txt
# Save run settings
with open(save_dir / 'hyp.yaml', 'w') as f: # 打开路径save_dir下的hyp.yaml可写文件
yaml.dump(hyp, f, sort_keys=False) # 使用dump函数将hyp写入文件f中
with open(save_dir / 'opt.yaml', 'w') as f: # 打开路径save_dir下的opt.yaml可写文件
yaml.dump(vars(opt), f, sort_keys=False) # 将命令行参数配置 opt 以相同的方式保存为 YAML 文件
# Configure
plots = not opt.evolve # create plots,判断是否创建图表
cuda = device.type != 'cpu' # 判断cuda是否存在
init_seeds(2 + rank) # 随机数种子
with open(opt.data) as f: # 打开sand.yaml文件
data_dict = yaml.load(f, Loader=yaml.SafeLoader) # data dict，将打开的f文件的数据加载到data_dict中
is_coco = opt.data.endswith('coco.yaml') # 检查 opt.data 文件的路径是否以 'coco.yaml' 结尾，从而确定是否正在使用 COCO 数据集的配置。
# Logging- Doing this before checking the dataset. Might update data_dict
loggers = {'wandb': None} # loggers dict
if rank in [-1, 0]:
opt.hyp = hyp # add hyperparameters
run_id = torch.load(weights).get('wandb_id') if weights.endswith('.pt') and os.path.isfile(weights) else None
wandb_logger = WandbLogger(opt, Path(opt.save_dir).stem, run_id, data_dict)
loggers['wandb'] = wandb_logger.wandb
data_dict = wandb_logger.data_dict
if wandb_logger.wandb:
weights, epochs, hyp = opt.weights, opt.epochs, opt.hyp # WandbLogger might update weights, epochs if resuming
nc = 1 if opt.single_cls else int(data_dict['nc']) # number of classes
names = ['item'] if opt.single_cls and len(data_dict['names']) != 1 else data_dict['names'] # class names
assert len(names) == nc, '%g names found for nc=%g dataset in %s' % (len(names), nc, opt.data) # check
# Model
pretrained = weights.endswith('.pt') # 判断权重是否以.pt结尾
if pretrained: # 预训练
with torch_distributed_zero_first(rank): # 检查是否有权重文件，如果没有，尝试下载
attempt_download(weights) # download if not found locally
ckpt = torch.load(weights, map_location=device) # load checkpoint，加载权重文件
model = Model(opt.cfg or ckpt['model'].yaml, ch=3, nc=nc, anchors=hyp.get('anchors')).to(device) # model模块
exclude = ['anchor'] if (opt.cfg or hyp.get('anchors')) and not opt.resume else [] # exclude keys
state_dict = ckpt['model'].float().state_dict() # to FP32
state_dict = intersect_dicts(state_dict, model.state_dict(), exclude=exclude) # intersect
model.load_state_dict(state_dict, strict=False) # load
logger.info('Transferred %g/%g items from %s' % (len(state_dict), len(model.state_dict()), weights)) # report
else:
model = Model(opt.cfg, ch=3, nc=nc, anchors=hyp.get('anchors')).to(device) # create
with torch_distributed_zero_first(rank): # 分布式训练使用
check_dataset(data_dict) # check
train_path = data_dict['train'] # 获取训练集、测试集图片路径
test_path = data_dict['val']
# Freeze
freeze = [] # parameter names to freeze (full or partial)，冻结模型的部分参数
for k, v in model.named_parameters():
v.requires_grad = True # train all layers，对所有层进行训练
if any(x in k for x in freeze):
print('freezing %s' % k) # 打印冻结的参数
v.requires_grad = False # 表示这些参数在训练过程中不会更新
# Optimizer
"""
nbs为模拟的batch_size;
就比如默认的话上面设置的opt.batch_size为16,这个nbs就为64，
也就是模型梯度累积了64/16=4(accumulate)次之后
再更新一次模型，变相的扩大了batch_size
"""
nbs = 64 # nominal batch size
accumulate = max(round(nbs / total_batch_size), 1) # accumulate loss before optimizing
hyp['weight_decay'] *= total_batch_size * accumulate / nbs # scale weight_decay,避免过拟合
logger.info(f"Scaled weight_decay = {hyp['weight_decay']}") # 日志记录
pg0, pg1, pg2 = [], [], [] # optimizer parameter groups，
for k, v in model.named_modules(): # 遍历所有层
if hasattr(v, 'bias') and isinstance(v.bias, nn.Parameter):
# 检查一个模块v是否具有bias属性，并且该属性是否是nn.Parameter类型。如果满足这两个条件，说明模块的偏置参数是需要被优化的。
pg2.append(v.bias) # biases，如果以上两个条件都满足，代码将该模块的偏置参数v.bias添加到pg2列表中
if isinstance(v, nn.BatchNorm2d):
# 这行代码检查一个模块v是否是nn.BatchNorm2d类型
pg0.append(v.weight) # no decay，如果满足条件，将该模块的weight参数添加到pg0列表中。
elif hasattr(v, 'weight') and isinstance(v.weight, nn.Parameter):
# 检查一个模块v是否具有weight属性，并且该属性是否是nn.Parameter类型
pg1.append(v.weight) # apply decay，将该模块的权重参数v.weight添加到pg1列表中
if opt.adam:
# 若opt.adam为true，则采用adam优化器，pg0为需要优化的数组，lr0为学习率，betas是一组用于计算梯度和梯度平方的超参数
optimizer = optim.Adam(pg0, lr=hyp['lr0'], betas=(hyp['momentum'], 0.999)) # adjust beta1 to momentum
else:
# 否则采用SGD优化器，momentum是动量参数，nesterov表示是否使用Nesterov动量。
optimizer = optim.SGD(pg0, lr=hyp['lr0'], momentum=hyp['momentum'], nesterov=True)
optimizer.add_param_group({'params': pg1, 'weight_decay': hyp['weight_decay']}) # add pg1 with weight_decay
optimizer.add_param_group({'params': pg2}) # add pg2 (biases)
logger.info('Optimizer groups: %g .bias, %g conv.weight, %g other' % (len(pg2), len(pg1), len(pg0)))
del pg0, pg1, pg2 # 删除临时变量，释放内存
# Scheduler https://arxiv.org/pdf/1812.01187.pdf
# https://pytorch.org/docs/stable/_modules/torch/optim/lr_scheduler.html#OneCycleLR
if opt.linear_lr: # 若为真，则选用线性学习率调度器
lf = lambda x: (1 - x / (epochs - 1)) * (1.0 - hyp['lrf']) + hyp['lrf'] # linear
else: # 否则选用余弦退火式调度器
lf = one_cycle(1, hyp['lrf'], epochs) # cosine 1->hyp['lrf']
scheduler = lr_scheduler.LambdaLR(optimizer, lr_lambda=lf) # 创建scheduler的学习率调度器对象，会根据训练的进程更新优化器的学习率。
# plot_lr_scheduler(optimizer, scheduler, epochs)
# EMA
ema = ModelEMA(model) if rank in [-1, 0] else None # 指数移动平均对象
# Resume：从断点中恢复训练
start_epoch, best_fitness = 0, 0.0
if pretrained:
# Optimizer
# 加载优化器与best_fitness
if ckpt['optimizer'] is not None:
optimizer.load_state_dict(ckpt['optimizer']) # 恢复之前保存的优化器状态
best_fitness = ckpt['best_fitness']
# EMA，恢复EMA状态
if ema and ckpt.get('ema'):
ema.ema.load_state_dict(ckpt['ema'].float().state_dict())
ema.updates = ckpt['updates']
# Results，恢复训练结果,并保存到文件中
if ckpt.get('training_results') is not None:
results_file.write_text(ckpt['training_results']) # write results.txt
# Epochs
start_epoch = ckpt['epoch'] + 1 # 加载训练的轮次，当训练完毕后，epoch为-1
if opt.resume: # 若resume为真，则检查start_epoch是否大于0，若不大于0,则输出字符串
assert start_epoch > 0, '%s training to %g epochs is finished, nothing to resume.' % (weights, epochs)
if epochs < start_epoch:
logger.info('%s has been trained for %g epochs. Fine-tuning for %g additional epochs.' %
(weights, ckpt['epoch'], epochs)) # 指出已经训练的周期数
epochs += ckpt['epoch'] # finetune additional epochs，更新训练周期数
del ckpt, state_dict # 清理内存空间
# Image sizes
gs = max(int(model.stride.max()), 32) # grid size (max stride) # gs为最大步幅的整数值：卷积操作每次移动的像素数
nl = model.model[-1].nl # number of detection layers (used for scaling hyp['obj']) # 获取模型最后一个检测层的数量
imgsz, imgsz_test = [check_img_size(x, gs) for x in opt.img_size] # verify imgsz are gs-multiples
# 检测opt.img_size是否是gs的整数倍，imgsz表示训练图像，imgsz_test表示测试图像
# DP mode，分布式训练
if cuda and rank == -1 and torch.cuda.device_count() > 1:
model = torch.nn.DataParallel(model)
# SyncBatchNorm，同步批归一化
if opt.sync_bn and cuda and rank != -1:
model = torch.nn.SyncBatchNorm.convert_sync_batchnorm(model).to(device)
logger.info('Using SyncBatchNorm()')
# Trainloader，创建训练集
dataloader, dataset = create_dataloader(train_path, imgsz, batch_size, gs, opt,
hyp=hyp, augment=True, cache=opt.cache_images, rect=opt.rect, rank=rank,
world_size=opt.world_size, workers=opt.workers,
image_weights=opt.image_weights, quad=opt.quad, prefix=colorstr('train: '))
mlc = np.concatenate(dataset.labels, 0)[:, 0].max() # max label class，获取标签的最大类别值
nb = len(dataloader) # number of batches，训练数据集被分成多少批次
assert mlc < nc, 'Label class %g exceeds nc=%g in %s. Possible class labels are 0-%g' % (mlc, nc, opt.data, nc - 1)
# 与类别数比较，若大于类别数代表有问题
# Process 0
if rank in [-1, 0]:
testloader = create_dataloader(test_path, imgsz_test, batch_size * 2, gs, opt, # testloader，验证集
hyp=hyp, cache=opt.cache_images and not opt.notest, rect=True, rank=-1,
world_size=opt.world_size, workers=opt.workers,
pad=0.5, prefix=colorstr('val: '))[0]
if not opt.resume:
labels = np.concatenate(dataset.labels, 0) # 所有样本标签组成一个numpy数组
c = torch.tensor(labels[:, 0]) # classes，标签中提取类别信息
# cf = torch.bincount(c.long(), minlength=nc) + 1. # frequency
# model._initialize_biases(cf.to(device))
if plots:
plot_labels(labels, names, save_dir, loggers) # 如果plots为真，则会绘制类别标签的分布图以及其他可视化图表，并保存
if tb_writer:
tb_writer.add_histogram('classes', c, 0) # 若tb_writer为真，则将直方图添加到TensorBoard
# Anchors
if not opt.noautoanchor: # noautoanchor未被调用，则默认为false
check_anchors(dataset, model=model, thr=hyp['anchor_t'], imgsz=imgsz) # 根据数据集和特征图自动调整锚框的尺寸
model.half().float() # pre-reduce anchor precision
# DDP mode，分布式训练使用
if cuda and rank != -1:
model = DDP(model, device_ids=[opt.local_rank], output_device=opt.local_rank,
# nn.MultiheadAttention incompatibility with DDP https://github.com/pytorch/pytorch/issues/26698
find_unused_parameters=any(isinstance(layer, nn.MultiheadAttention) for layer in model.modules()))
# Model parameters
# 这些超参数是损失函数的系数，适应不同训练数据和任务
hyp['box'] *= 3. / nl # scale to layers，nl为检测层的数量
hyp['cls'] *= nc / 80. * 3. / nl # scale to classes and layers，nc表示类别数
hyp['obj'] *= (imgsz / 640) ** 2 * 3. / nl # scale to image size and layers
hyp['label_smoothing'] = opt.label_smoothing # 用不到
model.nc = nc # attach number of classes to model，表示将类别数量附加到模型上
model.hyp = hyp # attach hyperparameters to model，表示将超参数附加到模型上
model.gr = 1.0 # iou loss ratio (obj_loss = 1.0 or iou)，用于控制是否使用IOU损失，iou表示交并比
model.class_weights = labels_to_class_weights(dataset.labels, nc).to(device) * nc
# attach class weights，model.class_weights 被设置为类别权重，这些权重用于调整不同类别在损失计算中的贡献
model.names = names # 表示将类别名称列表附加到模型上
# Start training
t0 = time.time() # 记录训练一轮的时间
# 获取热身阶段的迭代次数
nw = max(round(hyp['warmup_epochs'] * nb), 1000) # number of warmup iterations, max(3 epochs, 1k iterations)
# nw = min(nw, (epochs - start_epoch) / 2 * nb) # limit warmup to < 1/2 of training
maps = np.zeros(nc) # mAP per class，包含每个类别的MAP,MAP:精确率和召回率曲线下的面积
results = (0, 0, 0, 0, 0, 0, 0) # P, R, mAP@.5, mAP@.5-.95, val_loss(box, obj, cls)，包含不同指标的元组
"""
设置学习率衰减所进行到的轮次，
目的是打断训练后，--resume接着训练也能正常的衔接之前的训练进行学习率衰减
"""
scheduler.last_epoch = start_epoch - 1 # do not move
scaler = amp.GradScaler(enabled=cuda) # 缩放梯度
compute_loss = ComputeLoss(model) # init loss class，计算损失函数
logger.info(f'Image sizes {imgsz} train, {imgsz_test} test\n'
f'Using {dataloader.num_workers} dataloader workers\n'
f'Logging results to {save_dir}\n'
f'Starting training for {epochs} epochs...') # 加载训练的基本信息，--图像尺寸等
for epoch in range(start_epoch, epochs): # epoch ------------------------------------------------------------------
model.train() # 训练
# Update image weights (optional)，更新权重
if opt.image_weights:
# Generate indices
if rank in [-1, 0]:
cw = model.class_weights.cpu().numpy() * (1 - maps) ** 2 / nc # class weights，计算每个类别的权重
iw = labels_to_image_weights(dataset.labels, nc=nc, class_weights=cw) # image weights，根据类别权重计算图像权重
dataset.indices = random.choices(range(dataset.n), weights=iw, k=dataset.n) # rand weighted idx，根据iw对数据集随机采样
# Broadcast if DDP，分布式训练使用
if rank != -1:
indices = (torch.tensor(dataset.indices) if rank == 0 else torch.zeros(dataset.n)).int()
dist.broadcast(indices, 0)
if rank != 0:
dataset.indices = indices.cpu().numpy()
# Update mosaic border
# 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(4, device=device) # mean losses,保存平均损失，维度为4
if rank != -1:
dataloader.sampler.set_epoch(epoch) # 分布式训练
pbar = enumerate(dataloader) # 遍历每个批次数据
# 打印训练批次的表头信息
logger.info(('\n' + '%10s' * 8) % ('Epoch', 'gpu_mem', 'box', 'obj', 'cls', 'total', 'labels', 'img_size'))
if rank in [-1, 0]:
pbar = tqdm(pbar, total=nb) # progress bar，显示训练进度条，nb:批次的总数量
optimizer.zero_grad() # 优化器的梯度置零
# 遍历每个批次
for i, (imgs, targets, paths, _) in pbar: # batch -------------------------------------------------------------
ni = i + nb * epoch # number integrated batches (since train start)
imgs = imgs.to(device, non_blocking=True).float() / 255.0 # uint8 to float32, 0-255 to 0.0-1.0
# Warmup,用于学习率的调整，使模型在训练初期能更快适应数据，然后用稳定的学习率进行训练
if ni <= nw: # ni表示当前迭代的次数，nw代表预热的迭代次数
xi = [0, nw] # x interp，xi是一个用于插值的数组
# model.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 / total_batch_size]).round()) # 累积梯度的迭代次数,多少次迭代后进行一次梯度更新
for j, x in enumerate(optimizer.param_groups): # 循环遍历优化器中的参数
# 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 == 2 else 0.0, x['initial_lr'] * lf(epoch)]) # 学习率的调整
if 'momentum' in x: # 若当前参数包含动量值，则将动量值逐渐增加到正常值
x['momentum'] = np.interp(ni, xi, [hyp['warmup_momentum'], hyp['momentum']])
# Multi-scale，多尺度训练模式，根据输入的照片不同进行随机尺度变换
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 = F.interpolate(imgs, size=ns, mode='bilinear', align_corners=False)
# Forward，前向传播
with amp.autocast(enabled=cuda):
pred = model(imgs) # forward
loss, loss_items = compute_loss(pred, targets.to(device)) # loss scaled by batch_size
if rank != -1:
loss *= opt.world_size # gradient averaged between devices in DDP mode
if opt.quad:
loss *= 4.
# Backward，反向传播
scaler.scale(loss).backward()
# Optimize，模型优化，包括梯度累计，梯度更新，权重更新，EMA更新
if ni % accumulate == 0:
scaler.step(optimizer) # optimizer.step
scaler.update()
optimizer.zero_grad()
if ema:
ema.update(model)
# Print，打印训练过程的状态信息
if rank in [-1, 0]:
mloss = (mloss * i + loss_items) / (i + 1) # update mean losses
mem = '%.3gG' % (torch.cuda.memory_reserved() / 1E9 if torch.cuda.is_available() else 0) # (GB)，计算内存占用
s = ('%10s' * 2 + '%10.4g' * 6) % (
'%g/%g' % (epoch, epochs - 1), mem, *mloss, targets.shape[0], imgs.shape[-1])
pbar.set_description(s) # 更新进度条的描述信息
# Plot
if plots and ni < 3: # plots是一个bool值：是否可视化，ni为当前迭代次数
f = save_dir / f'train_batch{ni}.jpg' # filename，保存一个用于保存图像的文件
Thread(target=plot_images, args=(imgs, targets, paths, f), daemon=True).start() # 绘制图像并保存至文件中
# if tb_writer:
# tb_writer.add_image(f, result, dataformats='HWC', global_step=epoch)
# tb_writer.add_graph(torch.jit.trace(model, imgs, strict=False), []) # add model graph
elif plots and ni == 10 and wandb_logger.wandb: # wandb_logger.wandb存在，会在WandB中记录图像
wandb_logger.log({"Mosaics": [wandb_logger.wandb.Image(str(x), caption=x.name) for x in
save_dir.glob('train*.jpg') if x.exists()]})
# end batch ------------------------------------------------------------------------------------------------
# end epoch ----------------------------------------------------------------------------------------------------
# Scheduler，在tensorboard中记录学习率的变化
lr = [x['lr'] for x in optimizer.param_groups] # for tensorboard
scheduler.step() # 调整学习率
# DDP process 0 or single-GPU
if rank in [-1, 0]:
# mAP
# 给ema更新信息
ema.update_attr(model, include=['yaml', 'nc', 'hyp', 'gr', 'names', 'stride', 'class_weights'])
final_epoch = epoch + 1 == epochs # 判断是否是最后一轮
if not opt.notest or final_epoch: # Calculate mAP，若不是opt.notest（测试）或处于最后一轮
wandb_logger.current_epoch = epoch + 1 # 记录当前轮数
results, maps, times = test.test(data_dict,
batch_size=batch_size * 2,
imgsz=imgsz_test,
model=ema.ema,
single_cls=opt.single_cls,
dataloader=testloader,
save_dir=save_dir,
verbose=nc < 50 and final_epoch,
plots=plots and final_epoch,
wandb_logger=wandb_logger,
compute_loss=compute_loss,
is_coco=is_coco)
# Write
with open(results_file, 'a') as f: # 保存训练结果的状态信息
f.write(s + '%10.4g' * 7 % results + '\n') # append metrics, val_loss
if len(opt.name) and opt.bucket:
os.system('gsutil cp %s gs://%s/results/results%s.txt' % (results_file, opt.bucket, opt.name))
# Log，tags 是用于记录到 TensorBoard 和 WandB 中的标签列表
tags = ['train/box_loss', 'train/obj_loss', 'train/cls_loss', # train loss
'metrics/precision', 'metrics/recall', 'metrics/mAP_0.5', 'metrics/mAP_0.5:0.95',
'val/box_loss', 'val/obj_loss', 'val/cls_loss', # val loss
'x/lr0', 'x/lr1', 'x/lr2'] # params
for x, tag in zip(list(mloss[:-1]) + list(results) + lr, tags):
if tb_writer:
tb_writer.add_scalar(tag, x, epoch) # tensorboard
if wandb_logger.wandb:
wandb_logger.log({tag: x}) # W&B
# Update best mAP
# 拟合参数，记录综合参数
fi = fitness(np.array(results).reshape(1, -1)) # weighted combination of [P, R, mAP@.5, mAP@.5-.95]
if fi > best_fitness:
best_fitness = fi # 记录最佳参数
wandb_logger.end_epoch(best_result=best_fitness == fi) # 结束训练，并且将fi保存至wandb
# Save model
if (not opt.nosave) or (final_epoch and not opt.evolve): # if save,是否需要保存模型
ckpt = {'epoch': epoch,
'best_fitness': best_fitness,
'training_results': results_file.read_text(),
'model': deepcopy(model.module if is_parallel(model) else model).half(),
'ema': deepcopy(ema.ema).half(),
'updates': ema.updates,
'optimizer': optimizer.state_dict(),
'wandb_id': wandb_logger.wandb_run.id if wandb_logger.wandb else None}
# Save last, best and delete
torch.save(ckpt, last) # 将ckpt中文件保存至last文件中
if best_fitness == fi: # 若当前性能为最佳性能，则将其保存为best文件
torch.save(ckpt, best)
if wandb_logger.wandb: # 是否使用wandb记录信息
if ((epoch + 1) % opt.save_period == 0 and not final_epoch) and opt.save_period != -1:
wandb_logger.log_model(
last.parent, opt, epoch, fi, best_model=best_fitness == fi)
del ckpt
# end epoch ----------------------------------------------------------------------------------------------------
# end training
if rank in [-1, 0]:
# Plots
if plots:
plot_results(save_dir=save_dir) # save as results.png，是否绘制结果图
if wandb_logger.wandb:
files = ['results.png', 'confusion_matrix.png', *[f'{x}_curve.png' for x in ('F1', 'PR', 'P', 'R')]]
# files为一个列表包含了要上传但wandb的图像文件名
wandb_logger.log({"Results": [wandb_logger.wandb.Image(str(save_dir / f), caption=f) for f in files
if (save_dir / f).exists()]})
# caption=f 指定图像的标题为文件名，对每个图像文件名进行循环遍历，并仅处理存在的文件。
# Test best.pt
# 记录训练过程的总耗时，以及完成的 epoch 数量和所用时间
logger.info('%g epochs completed in %.3f hours.\n' % (epoch - start_epoch + 1, (time.time() - t0) / 3600))
if opt.data.endswith('coco.yaml') and nc == 80: # if COCO，检查数据集是否是 COCO 数据集，并且类别数量是否为 80
for m in (last, best) if best.exists() else (last): # speed, mAP tests，是否存在best模型。若存在，会测last和best，否则只测last
results, _, _ = test.test(opt.data,
batch_size=batch_size * 2,
imgsz=imgsz_test,
conf_thres=0.001,
iou_thres=0.7,
model=attempt_load(m, device).half(),
single_cls=opt.single_cls,
dataloader=testloader,
save_dir=save_dir,
save_json=True,
plots=False,
is_coco=is_coco)
# Strip optimizers
final = best if best.exists() else last # final model，最终使用的模型是best模型（如果存在），否则是last模型。
for f in last, best:
if f.exists():
strip_optimizer(f) # strip optimizers，用于去除模型中的优化器参数
if opt.bucket:
os.system(f'gsutil cp {final} gs://{opt.bucket}/weights') # upload，若设置了opt.bucket，将最终模型文件上传到云存储。
if wandb_logger.wandb and not opt.evolve: # Log the stripped model
wandb_logger.wandb.log_artifact(str(final), type='model',
name='run_' + wandb_logger.wandb_run.id + '_model',
aliases=['last', 'best', 'stripped'])
wandb_logger.finish_run() # 结束WandB记录的运行
else:
dist.destroy_process_group()
torch.cuda.empty_cache() # 清理GPU内存中的缓存，以释放空间
return results
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--weights', type=str, default='weights/yolov5s.pt', help='initial weights path')
parser.add_argument('--cfg', type=str, default='models/sands.yaml', help='model.yaml path')
parser.add_argument('--data', type=str, default='data/sand.yaml', help='data.yaml path')
parser.add_argument('--hyp', type=str, default='data/hyp.scratch.yaml', help='hyperparameters path')
parser.add_argument('--epochs', type=int, default=100)
parser.add_argument('--batch-size', type=int, default=5, help='total batch size for all GPUs')
parser.add_argument('--img-size', nargs='+', type=int, default=[640, 640], help='[train, test] image sizes')
parser.add_argument('--rect', action='store_true', help='rectangular training')
parser.add_argument('--resume', nargs='?', const=True, default=False, help='resume most recent training')
parser.add_argument('--nosave', action='store_true', help='only save final checkpoint')
parser.add_argument('--notest', action='store_true', help='only test final epoch')
parser.add_argument('--noautoanchor', action='store_true', help='disable autoanchor check')
parser.add_argument('--evolve', action='store_true', help='evolve hyperparameters')
parser.add_argument('--bucket', type=str, default='', help='gsutil bucket')
parser.add_argument('--cache-images', action='store_true', help='cache images for faster training')
parser.add_argument('--image-weights', action='store_true', help='use weighted image selection for training')
parser.add_argument('--device', default='', help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
parser.add_argument('--multi-scale', action='store_true', help='vary img-size +/- 50%%')
parser.add_argument('--single-cls', action='store_true', help='train multi-class data as single-class')
parser.add_argument('--adam', action='store_true', help='use torch.optim.Adam() optimizer')
parser.add_argument('--sync-bn', action='store_true', help='use SyncBatchNorm, only available in DDP mode')
parser.add_argument('--local_rank', type=int, default=-1, help='DDP parameter, do not modify')
parser.add_argument('--workers', type=int, default=8, help='maximum number of dataloader workers')
parser.add_argument('--project', default='runs/train', help='save to project/name')
parser.add_argument('--entity', default=None, help='W&B entity')
parser.add_argument('--name', default='exp', help='save to project/name')
parser.add_argument('--exist-ok', action='store_true', help='existing project/name ok, do not increment')
parser.add_argument('--quad', action='store_true', help='quad dataloader')
parser.add_argument('--linear-lr', action='store_true', help='linear LR')
parser.add_argument('--label-smoothing', type=float, default=0.0, help='Label smoothing epsilon')
parser.add_argument('--upload_dataset', action='store_true', help='Upload dataset as W&B artifact table')
parser.add_argument('--bbox_interval', type=int, default=-1, help='Set bounding-box image logging interval for W&B')
parser.add_argument('--save_period', type=int, default=-1, help='Log model after every "save_period" epoch')
parser.add_argument('--artifact_alias', type=str, default="latest", help='version of dataset artifact to be used')
opt = parser.parse_args()
# Set DDP variables
opt.world_size = int(os.environ['WORLD_SIZE']) if 'WORLD_SIZE' in os.environ else 1 # 设置参数值
opt.global_rank = int(os.environ['RANK']) if 'RANK' in os.environ else -1
set_logging(opt.global_rank)
if opt.global_rank in [-1, 0]: # 判断global_rank是否为-1或0
check_git_status()
check_requirements() # 检查依赖是否安装
# Resume
wandb_run = check_wandb_resume(opt)
if opt.resume and not wandb_run: # resume an interrupted run
ckpt = opt.resume if isinstance(opt.resume, str) else get_latest_run() # specified or most recent path
assert os.path.isfile(ckpt), 'ERROR: --resume checkpoint does not exist'
apriori = opt.global_rank, opt.local_rank
with open(Path(ckpt).parent.parent / 'opt.yaml') as f:
opt = argparse.Namespace(**yaml.load(f, Loader=yaml.SafeLoader)) # replace
opt.cfg, opt.weights, opt.resume, opt.batch_size, opt.global_rank, opt.local_rank = '', ckpt, True, opt.total_batch_size, *apriori # reinstate
logger.info('Resuming training from %s' % ckpt)
else:
# opt.hyp = opt.hyp or ('hyp.finetune.yaml' if opt.weights else 'hyp.scratch.yaml')
opt.data, opt.cfg, opt.hyp = check_file(opt.data), check_file(opt.cfg), check_file(opt.hyp) # check files
# 检查data,cfg,hyp文件是否存在，hyp为超参数
assert len(opt.cfg) or len(opt.weights), 'either --cfg or --weights must be specified'
# 使用断言来检查 opt.cfg 和 opt.weights 属性是否至少有一个不为空，因为在训练时要么提供配置文件 --cfg，要么提供权重文件 --weights
opt.img_size.extend([opt.img_size[-1]] * (2 - len(opt.img_size))) # extend to 2 sizes (train, test)
opt.name = 'evolve' if opt.evolve else opt.name
# 如果 opt.evolve 为真，则将 opt.name 设置为 'evolve'，否则保持不变。--未使用evolve选项
opt.save_dir = increment_path(Path(opt.project) / opt.name, exist_ok=opt.exist_ok | opt.evolve) # increment run
# 增量路径--runs/train/exp_,如果设置了 opt.evolve 或 opt.exist_ok，则使用 increment_path 函数以避免目录重名。
# DDP mode
opt.total_batch_size = opt.batch_size # 选择使用的设备是cpu还是gpu
device = select_device(opt.device, batch_size=opt.batch_size)
if opt.local_rank != -1: # 是否进行分布式训练
assert torch.cuda.device_count() > opt.local_rank
torch.cuda.set_device(opt.local_rank)
device = torch.device('cuda', opt.local_rank)
dist.init_process_group(backend='nccl', init_method='env://') # distributed backend
assert opt.batch_size % opt.world_size == 0, '--batch-size must be multiple of CUDA device count'
opt.batch_size = opt.total_batch_size // opt.world_size
# Hyperparameters
with open(opt.hyp) as f: # 打开指定的超参数文件
hyp = yaml.load(f, Loader=yaml.SafeLoader) # load hyps # 从yaml模块中的load函数加载超参数，yaml.SafeLoader 被用于安全地加载 YAML 数据
# Train
logger.info(opt)
if not opt.evolve:
tb_writer = None # init loggers，初始化 TensorBoard 记录器
if opt.global_rank in [-1, 0]:
prefix = colorstr('tensorboard: ')
logger.info(f"{prefix}Start with 'tensorboard --logdir {opt.project}', view at http://localhost:6006/")
# 在日志中输出 TensorBoard 的启动提示信息，告诉您如何在本地查看 TensorBoard 的可视化结果。
tb_writer = SummaryWriter(opt.save_dir)
# Tensorboard，创建一个 TensorBoard 记录器，保存在指定的 opt.save_dir 目录下--runs/train/exp_
train(hyp, opt, device, tb_writer) # 训练
# 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)
assert opt.local_rank == -1, 'DDP mode not implemented for --evolve'
opt.notest, opt.nosave = True, True # only test/save final epoch
# ei = [isinstance(x, (int, float)) for x in hyp.values()] # evolvable indices
yaml_file = Path(opt.save_dir) / 'hyp_evolved.yaml' # save best result here
if opt.bucket:
os.system('gsutil cp gs://%s/evolve.txt .' % opt.bucket) # download evolve.txt if exists
for _ in range(300): # generations to evolve
if Path('evolve.txt').exists(): # if evolve.txt exists: select best hyps and mutate
# Select parent(s)
parent = 'single' # parent selection method: 'single' or 'weighted'
x = np.loadtxt('evolve.txt', ndmin=2)
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() # weights
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([x[0] for x in meta.values()]) # 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)
# Write mutation results
print_mutation(hyp.copy(), results, yaml_file, opt.bucket)
# Plot results
plot_evolution(yaml_file)
print(f'Hyperparameter evolution complete. Best results saved as: {yaml_file}\n'
f'Command to train a new model with these hyperparameters: $ python train.py --hyp {yaml_file}')
10-29
6929
6929
08-18
7922
7922
05-08
1万+
1万+
07-25
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
