def __init__(self, weight_path, resume, gpu_id, accumulate, fp_16):
init_seeds(0)
self.fp_16 = fp_16
self.device = gpu.select_device(gpu_id)
self.start_epoch = 0
self.best_mAP = 0.
self.accumulate = accumulate
self.epochs = cfg.TRAIN["EPOCHS"]
self.weight_path = weight_path
self.multi_scale_train = cfg.TRAIN["MULTI_SCALE_TRAIN"]
self.train_dataset = data.Build_Dataset(anno_file_type="train", img_size=cfg.TRAIN["TRAIN_IMG_SIZE"])
print('train img size is {}'.format(cfg.TRAIN["TRAIN_IMG_SIZE"]))
self.train_dataloader = DataLoader(self.train_dataset,
batch_size=cfg.TRAIN["BATCH_SIZE"],
num_workers=cfg.TRAIN["NUMBER_WORKERS"],
shuffle=True, pin_memory=True
)
self.yolov4 = Build_Model().to(self.device)
self.optimizer = optim.SGD(self.yolov4.parameters(), lr=cfg.TRAIN["LR_INIT"],
momentum=cfg.TRAIN["MOMENTUM"], weight_decay=cfg.TRAIN["WEIGHT_DECAY"])
self.criterion = YoloV4Loss(anchors=cfg.MODEL["ANCHORS"], strides=cfg.MODEL["STRIDES"],
iou_threshold_loss=cfg.TRAIN["IOU_THRESHOLD_LOSS"])
self.__load_model_weights(weight_path, resume)
self.scheduler = cosine_lr_scheduler.CosineDecayLR(self.optimizer,
T_max=self.epochs*len(self.train_dataloader),
lr_init=cfg.TRAIN["LR_INIT"],
lr_min=cfg.TRAIN["LR_END"],
warmup=cfg.TRAIN["WARMUP_EPOCHS"]*len(self.train_dataloader))
def __init__(self,
weight_path=None,
resume=False,
gpu_id=0,
accumulate=1,
fp_16=False):
init_seeds(0)
self.fp_16 = fp_16
self.device = gpu.select_device(gpu_id)
self.start_epoch = 0
self.best_mAP = 0.0
self.accumulate = accumulate
self.weight_path = weight_path
self.multi_scale_train = cfg.TRAIN["MULTI_SCALE_TRAIN"]
self.showatt = cfg.TRAIN["showatt"]
if self.multi_scale_train:
print("Using multi scales training")
else:
print("train img size is {}".format(cfg.TRAIN["TRAIN_IMG_SIZE"]))
self.train_dataset = data.Build_Dataset(
anno_file_type="train", img_size=cfg.TRAIN["TRAIN_IMG_SIZE"])
self.epochs = (cfg.TRAIN["YOLO_EPOCHS"] if cfg.MODEL_TYPE["TYPE"]
== "YOLOv4" else cfg.TRAIN["Mobilenet_YOLO_EPOCHS"])
self.eval_epoch = (30 if cfg.MODEL_TYPE["TYPE"] == "YOLOv4" else 50)
self.train_dataloader = DataLoader(
self.train_dataset,
batch_size=cfg.TRAIN["BATCH_SIZE"],
num_workers=cfg.TRAIN["NUMBER_WORKERS"],
shuffle=True,
pin_memory=True,
)
self.yolov4 = Build_Model(weight_path=weight_path,
resume=resume,
showatt=self.showatt).to(self.device)
self.optimizer = optim.SGD(
self.yolov4.parameters(),
lr=cfg.TRAIN["LR_INIT"],
momentum=cfg.TRAIN["MOMENTUM"],
weight_decay=cfg.TRAIN["WEIGHT_DECAY"],
)
self.criterion = YoloV4Loss(
anchors=cfg.MODEL["ANCHORS"],
strides=cfg.MODEL["STRIDES"],
iou_threshold_loss=cfg.TRAIN["IOU_THRESHOLD_LOSS"],
)
self.scheduler = cosine_lr_scheduler.CosineDecayLR(
self.optimizer,
T_max=self.epochs * len(self.train_dataloader),
lr_init=cfg.TRAIN["LR_INIT"],
lr_min=cfg.TRAIN["LR_END"],
warmup=cfg.TRAIN["WARMUP_EPOCHS"] * len(self.train_dataloader),
)
if resume:
self.__load_resume_weights(weight_path)
def __init__(self, weight_path, resume, gpu_id, vis, mode=None):
init_seeds(0)
self.device = gpu.select_device(gpu_id)
self.start_epoch = 0
self.best_mAP = 0.
self.epochs = cfg.epoch = 100
self.weight_path = weight_path
self.resume = resume
self.mode = mode
self.multi_scale_train = cfg.MULTI_SCALE_TRAIN
print('Loading Datasets...')
self.train_dataset = PASCALVOC(img_size=cfg.img_size,
root=cfg.root,
image_sets=cfg.train_sets,
phase='trainval',
mean=cfg.means,
std=cfg.std)
self.val_dataset = PASCALVOC(img_size=cfg.img_size,
root=cfg.root,
image_sets=cfg.test_sets,
phase='test',
mean=cfg.means,
std=cfg.std)
self.train_dataloader = DataLoader(self.train_dataset,
batch_size=cfg.batch_size,
num_workers=cfg.workers,
collate_fn=detection_collate,
shuffle=True)
if vis:
ViewDatasets(self.train_dataloader)
self.SSD = SSD(num_classes=cfg.num_classes,
num_blocks=cfg.mbox,
top_k=cfg.top_k,
conf_thresh=cfg.conf_thresh,
nms_thresh=cfg.nms_thresh,
variance=cfg.variance).to(self.device)
self.optimizer = optim.SGD(self.SSD.parameters(),
lr=cfg.init_lr,
momentum=cfg.momentum,
weight_decay=cfg.weight_decay)
self.criterion = SSD_loss(num_classes=cfg.num_classes,
variances=cfg.variance,
device=self.device)
self.__load_model_weights(self.weight_path, self.resume, self.mode)
self.scheduler = cosine_lr_scheduler.CosineDecayLR(
self.optimizer,
T_max=self.epochs * len(self.train_dataloader),
lr_init=cfg.init_lr,
lr_min=cfg.end_lr,
warmup=cfg.warmup_epoch * len(self.train_dataloader))
def __init__(self, log_dir, resume=False, fine_tune=False):
init_seeds(0)
if fine_tune:
self.__prepare_fine_tune()
self.fp_16 = cfg.FP16
self.device = gpu.select_device()
self.start_epoch = 0
self.best_mAP = 0.
self.accumulate = cfg.TRAIN.ACCUMULATE
self.log_dir = log_dir
self.weight_path = "/content/drive/MyDrive/YOLO/weights/yolov4.weights"
self.multi_scale_train = cfg.TRAIN.MULTI_SCALE_TRAIN
if self.multi_scale_train:
print('Using multi scales training')
else:
print('train img size is {}'.format(cfg.TRAIN.TRAIN_IMG_SIZE))
self.train_dataset = data.Build_Train_Dataset(
anno_file=cfg.TRAIN.ANNO_FILE,
anno_file_type="train",
img_size=cfg.TRAIN.TRAIN_IMG_SIZE)
self.epochs = cfg.TRAIN.YOLO_EPOCHS if cfg.MODEL.MODEL_TYPE == 'YOLOv4' else cfg.TRAIN.Mobilenet_YOLO_EPOCHS
self.train_dataloader = DataLoader(
self.train_dataset,
batch_size=cfg.TRAIN.BATCH_SIZE // cfg.TRAIN.ACCUMULATE,
num_workers=cfg.TRAIN.NUMBER_WORKERS,
shuffle=True,
pin_memory=True)
self.yolov4 = Build_Model(
weight_path="/content/drive/MyDrive/YOLO/weights/yolov4.weights",
resume=resume)
self.yolov4 = self.yolov4.to(self.device)
self.optimizer = optim.SGD(self.yolov4.parameters(),
lr=cfg.TRAIN.LR_INIT,
momentum=cfg.TRAIN.MOMENTUM,
weight_decay=cfg.TRAIN.WEIGHT_DECAY)
self.criterion = YoloV4Loss(
anchors=cfg.MODEL.ANCHORS,
strides=cfg.MODEL.STRIDES,
iou_threshold_loss=cfg.TRAIN.IOU_THRESHOLD_LOSS)
self.scheduler = cosine_lr_scheduler.CosineDecayLR(
self.optimizer,
T_max=self.epochs * len(self.train_dataloader),
lr_init=cfg.TRAIN.LR_INIT,
lr_min=cfg.TRAIN.LR_END,
warmup=cfg.TRAIN.WARMUP_EPOCHS * len(self.train_dataloader))
if resume: self.__load_resume_weights()
if fine_tune: self.__load_best_weights()
def __init__(self, cfg, train_dir, valid_dir, weight_path, resume, gpu_id):
init_seeds(0)
self.cfg = cfg
self.train_dir = train_dir
self.device = gpu.select_device(gpu_id)
self.start_epoch = 0
self.best_mAP = 0.
self.epochs = cfg.TRAIN["EPOCHS"]
self.weight_path = weight_path
self.today = datetime.datetime.today().strftime('%y%m%d')
num_weights = len(glob.glob(f'./weight/{self.today}_*'))
self.save_weight_dir = f'./weight/{self.today}_{num_weights+1}'
os.mkdir(self.save_weight_dir)
self.multi_scale_train = cfg.TRAIN["MULTI_SCALE_TRAIN"]
# self.train_dataset = VocDataset(cfg=cfg, anno_file_type="train", img_size=cfg.TRAIN["TRAIN_IMG_SIZE"])
self.train_dataset = YoloDataset(cfg=cfg, data_dir=self.train_dir, img_size=cfg.TRAIN["TRAIN_IMG_SIZE"])
self.train_dataloader = DataLoader(self.train_dataset,
batch_size=cfg.TRAIN["BATCH_SIZE"],
num_workers=cfg.TRAIN["NUMBER_WORKERS"],
shuffle=True)
self.yolov3 = Yolov3(cfg=cfg).to(self.device)
self.optimizer = optim.SGD(self.yolov3.parameters(), lr=cfg.TRAIN["LR_INIT"],
momentum=cfg.TRAIN["MOMENTUM"], weight_decay=cfg.TRAIN["WEIGHT_DECAY"])
#self.optimizer = optim.Adam(self.yolov3.parameters(), lr = lr_init, weight_decay=0.9995)
self.criterion = YoloV3Loss(anchors=cfg.MODEL["ANCHORS"], strides=cfg.MODEL["STRIDES"],
iou_threshold_loss=cfg.TRAIN["IOU_THRESHOLD_LOSS"])
self.__load_model_weights(weight_path, resume)
self.scheduler = cosine_lr_scheduler.CosineDecayLR(self.optimizer,
T_max=self.epochs*len(self.train_dataloader),
lr_init=cfg.TRAIN["LR_INIT"],
lr_min=cfg.TRAIN["LR_END"],
warmup=cfg.TRAIN["WARMUP_EPOCHS"]*len(self.train_dataloader))
self.valid_dir = valid_dir
self.img_valid_dir = os.path.join(self.valid_dir, 'images')
self.anno_valid_dir = os.path.join(self.valid_dir, 'labels')
self.valid_dataset = YoloDataset(cfg=cfg, data_dir=self.valid_dir, img_size=cfg.EVAL["TEST_IMG_SIZE"])
self.valid_dataloader = DataLoader(self.valid_dataset,
batch_size=cfg.EVAL["BATCH_SIZE"],
num_workers=cfg.EVAL["NUMBER_WORKERS"],
shuffle=False)
self.evaluator = YoloEvaluator(self.yolov3, self.cfg, self.img_valid_dir, "eval", anno_dir=self.anno_valid_dir)
def __init__(self, weight_path, resume, gpu_id):
init_seeds(0)
self.device = gpu.select_device(gpu_id)
print(self.device)
self.start_epoch = 0
self.best_mAP = 0.
self.epochs = cfg.TRAIN["EPOCHS"]
self.weight_path = weight_path
self.multi_scale_train = cfg.TRAIN["MULTI_SCALE_TRAIN"]
if self.multi_scale_train: print('Using multi scales training')
else: print('train img size is {}'.format(cfg.TRAIN["TRAIN_IMG_SIZE"]))
self.train_dataset = data.Construct_Dataset(
anno_file_type="train", img_size=cfg.TRAIN["TRAIN_IMG_SIZE"])
self.train_dataloader = DataLoader(
self.train_dataset,
batch_size=cfg.TRAIN["BATCH_SIZE"],
num_workers=cfg.TRAIN["NUMBER_WORKERS"],
shuffle=True)
net_model = NPMMRDet()
if torch.cuda.device_count() > 1: ## multi GPUs
print("Let's use", torch.cuda.device_count(), "GPUs!")
net_model = torch.nn.DataParallel(net_model)
self.model = net_model.to(self.device)
elif torch.cuda.device_count() == 1:
self.model = net_model.to(self.device) ## Single GPU
self.optimizer = optim.SGD(self.model.parameters(),
lr=cfg.TRAIN["LR_INIT"],
momentum=cfg.TRAIN["MOMENTUM"],
weight_decay=cfg.TRAIN["WEIGHT_DECAY"])
self.criterion = Loss(
anchors=cfg.MODEL["ANCHORS"],
strides=cfg.MODEL["STRIDES"],
iou_threshold_loss=cfg.TRAIN["IOU_THRESHOLD_LOSS"])
self.__load_model_weights(weight_path, resume)
#self.__save_model_weights_best(160)
self.scheduler = cosine_lr_scheduler.CosineDecayLR(
self.optimizer,
T_max=self.epochs * len(self.train_dataloader),
lr_init=cfg.TRAIN["LR_INIT"],
lr_min=cfg.TRAIN["LR_END"],
warmup=cfg.TRAIN["WARMUP_EPOCHS"] * len(self.train_dataloader))
def __init__(self, model, weight_path, resume, gpu_id):
init_seeds(0)
self.model = model
self.device = gpu.select_device(gpu_id)
self.start_epoch = 0
self.best_mAP = 0.
self.epochs = cfg.TRAIN["EPOCHS"]
self.weight_path = weight_path
self.multi_scale_train = cfg.TRAIN["MULTI_SCALE_TRAIN"]
self.train_dataset = data.VocDataset(
anno_file_type="train", img_size=cfg.TRAIN["TRAIN_IMG_SIZE"])
self.train_dataloader = DataLoader(
self.train_dataset,
batch_size=cfg.TRAIN["BATCH_SIZE"],
num_workers=cfg.TRAIN["NUMBER_WORKERS"],
shuffle=True)
if self.model == 's':
self.yolov3 = Yolov3_S().to(self.device)
self.model_postfix = '_s'
elif self.model == 'l':
self.yolov3 = Yolov3_L().to(self.device)
self.model_postfix = '_l'
else:
self.yolov3 = Yolov3().to(self.device)
self.model_postfix = '_m'
# self.yolov3.apply(tools.weights_init_normal)
self.optimizer = optim.SGD(self.yolov3.parameters(),
lr=cfg.TRAIN["LR_INIT"],
momentum=cfg.TRAIN["MOMENTUM"],
weight_decay=cfg.TRAIN["WEIGHT_DECAY"])
#self.optimizer = optim.Adam(self.yolov3.parameters(), lr=cfg.TRAIN["LR_INIT"], weight_decay=0.9995)
self.criterion = YoloV3Loss(
anchors=cfg.MODEL["ANCHORS"],
strides=cfg.MODEL["STRIDES"],
iou_threshold_loss=cfg.TRAIN["IOU_THRESHOLD_LOSS"])
self.__load_model_weights(weight_path, resume)
self.scheduler = cosine_lr_scheduler.CosineDecayLR(
self.optimizer,
T_max=self.epochs * len(self.train_dataloader),
lr_init=cfg.TRAIN["LR_INIT"],
lr_min=cfg.TRAIN["LR_END"],
warmup=cfg.TRAIN["WARMUP_EPOCHS"] * len(self.train_dataloader))
init_method='tcp://127.0.0.1:9999', # distributed training init method
world_size=1, # number of nodes for distributed training
rank=0) # distributed training node rank
model = torch.nn.parallel.DistributedDataParallel(model, find_unused_parameters=True) # clw note: 多卡,在 amp.initialize()之后调用分布式代码 DistributedDataParallel否则报错
model.yolo_layers = model.module.yolo_layers # move yolo layer indices to top level
######
model.nc = nc
#### 阶梯学习率
#scheduler = lr_scheduler.MultiStepLR(optimizer, milestones=[round(total_epochs * x) for x in [0.8, 0.9]], gamma=0.1)
### 余弦学习率
# lf = lambda x: (1 + math.cos(x * math.pi / total_epochs)) / 2
# scheduler = lr_scheduler.LambdaLR(optimizer, lr_lambda=lf)
scheduler = cosine_lr_scheduler.CosineDecayLR(optimizer,
T_max=total_epochs * len(dataloader),
lr_init=lr0,
lr_min=lr0 * 0.01,
warmup=2 * len(dataloader))
# 4、训练
print('') # 换行
print('Starting training for %g epochs...' % total_epochs)
nb = len(dataloader)
mloss = torch.zeros(4).to(device) # mean losses
writer = SummaryWriter() # tensorboard --logdir=runs, view at http://localhost:6006/
prebias = start_epoch == 0
for epoch in range(start_epoch, total_epochs): # epoch ------------------------------
model.train() # 写在这里，是因为在一个epoch结束后，调用test.test()时，会调用 model.eval()
def __init__(self,
weight_path=None,
resume: bool = False,
gpu_id: int = 0,
accumulate: bool = True,
fp_16: bool = False):
# PYTHON HASH SEED
init_seeds(0)
# device
self.fp_16: bool = fp_16
self.device: torch.device = gpu.select_device(gpu_id)
self.start_epoch: int = 0
self.best_mAP: float = 0.0 # not sure why this is necessary...
self.accumulate: bool = accumulate
self.weight_path: Path = weight_path
self.multi_scale_train: bool = cfg.TRAIN["MULTI_SCALE_TRAIN"]
# Show attention modification?
self.showatt = cfg.TRAIN["showatt"]
# Multi-scale training status
if self.multi_scale_train:
print("Using multi scales training")
else:
print(f"train img size is {cfg.TRAIN['TRAIN_IMG_SIZE']}")
# Build Dataset using helper function.
self.train_dataset = data.Build_Dataset(
anno_file_type="train", img_size=cfg.TRAIN["TRAIN_IMG_SIZE"])
self.epochs = (cfg.TRAIN["YOLO_EPOCHS"] if cfg.MODEL_TYPE["TYPE"]
== "YOLOv4" else cfg.TRAIN["Mobilenet_YOLO_EPOCHS"])
self.eval_epoch = (30 if cfg.MODEL_TYPE["TYPE"] == "YOLOv4" else 50)
self.train_dataloader = DataLoader(
self.train_dataset,
batch_size=cfg.TRAIN["BATCH_SIZE"],
num_workers=cfg.TRAIN["NUMBER_WORKERS"],
shuffle=True,
pin_memory=True,
)
self.yolov4 = Build_Model(weight_path=weight_path,
resume=resume,
showatt=self.showatt).to(self.device)
self.optimizer = optim.SGD(
self.yolov4.parameters(),
lr=cfg.TRAIN["LR_INIT"],
momentum=cfg.TRAIN["MOMENTUM"],
weight_decay=cfg.TRAIN["WEIGHT_DECAY"],
)
self.criterion = YoloV4Loss(
anchors=cfg.MODEL["ANCHORS"],
strides=cfg.MODEL["STRIDES"],
iou_threshold_loss=cfg.TRAIN["IOU_THRESHOLD_LOSS"],
)
self.scheduler = cosine_lr_scheduler.CosineDecayLR(
self.optimizer,
T_max=self.epochs * len(self.train_dataloader),
lr_init=cfg.TRAIN["LR_INIT"],
lr_min=cfg.TRAIN["LR_END"],
warmup=cfg.TRAIN["WARMUP_EPOCHS"] * len(self.train_dataloader),
)
if resume:
self.__load_resume_weights(weight_path)
yolov4.parameters(),
lr=cfg.TRAIN["LR_INIT"],
momentum=cfg.TRAIN["MOMENTUM"],
weight_decay=cfg.TRAIN["WEIGHT_DECAY"],
)
criterion = YoloV4Loss(
anchors=cfg.MODEL["ANCHORS"],
strides=cfg.MODEL["STRIDES"],
iou_threshold_loss=cfg.TRAIN["IOU_THRESHOLD_LOSS"],
)
scheduler = cosine_lr_scheduler.CosineDecayLR(
optimizer,
T_max=epochs * len(train_dataloader),
lr_init=cfg.TRAIN["LR_INIT"],
lr_min=cfg.TRAIN["LR_END"],
warmup=cfg.TRAIN["WARMUP_EPOCHS"] * len(train_dataloader),
)
# Training
for epoch in range(start_epoch, epochs):
yolov4.train()
mloss = torch.zeros(4)
for i, data in enumerate(train_dataloader):
scheduler.step(
len(train_dataloader) / (cfg.TRAIN["BATCH_SIZE"]) * epoch + i)
imgs = data[0]
label_sbbox = data[1]
