Mixformer train代码解读

def run(settings):
    # Build dataloaders
    loader_train, loader_val = build_dataloaders(cfg, settings)

    # Create network
    net = build_mixformer(cfg)

    # wrap networks to distributed one
    settings.deep_sup = getattr(cfg.TRAIN, "DEEP_SUPERVISION", False)
    settings.distill = getattr(cfg.TRAIN, "DISTILL", False)
    settings.distill_loss_type = getattr(cfg.TRAIN, "DISTILL_LOSS_TYPE", "KL")
    # settings.save_every_epoch = True
    # Loss functions and Actors
    objective = {
   'giou': giou_loss, 'l1': l1_loss}
    loss_weight = {
   'giou': cfg.TRAIN.GIOU_WEIGHT, 'l1': cfg.TRAIN.L1_WEIGHT}
    actor = MixFormerActor(net=net, objective=objective, loss_weight=loss_weight, settings=settings)
		super().__init__(net, objective)
        self.loss_weight = loss_weight
        self.settings = settings
        self.bs = self.settings.batchsize  # batch size   8
        self.run_score_head = run_score_head   #false
        
    # Optimizer, parameters, and learning rates
    optimizer, lr_scheduler = get_optimizer_scheduler(net, cfg)
			#AdamW   
    use_amp = getattr(cfg.TRAIN, "AMP", False)
    trainer = LTRTrainer(actor, [loader_train, loader_val], optimizer, settings, lr_scheduler, use_amp=use_amp)
    		# Initialize statistics variables
        	# Initialize tensorboard

    # train process
    trainer.train(cfg.TRAIN.EPOCH, load_latest=True, fail_safe=True)

		 for epoch in range(self.epoch+1, max_epochs+1):
			self.train_epoch()
			self.cycle_dataset(loader)
			"""Do a cycle of training or validation."""
				for i, data in enumerate(loader, 1):  150/7500，这里的i是150，指的是当前运行的次数  7500是len(loader)
					# forward pass
					loss, stats = self.actor(data)
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
						class MixFormerActor(BaseActor):
							def __call__(self, data):
								out_dict = self.forward_pass(data, run_score_head=self.run_score_head)
									search_bboxes = box_xywh_to_xyxy(data['search_anno'][0].clone())    8  4
        							out_dict, _ = self.net(data['template_images'][0], data['template_images'][1], data['search_images'],run_score_head=run_score_head, gt_bboxes=search_bboxes) #  8 3 128 128    8 3 128 128   1 8 3 320 320
        					# out_dict: (B, N, C), outputs_coord: (1, B, N, C), target_query: (1, B, N, C)
-----------------------------------------------------------------------------------------------------------------------
        							    def forward(self, template, online_template, search, run_score_head=False, gt_bboxes=None):
        									template, search = self.backbone(template, online_template, search)
        									---->8 3 128 128          8 3 128 128        8 3 320 320
        									# Forward the corner head
        											def forward(self, template, online_template, search):   i：0 1 2
        												 template, online_template, search = getattr(self, f'stage{
     i}')(template, online_template, search)
**