Peeking into occluded joints: A novel framework for crowd pose estimation源码解读

    args = opt 		# 此处为解析参数，opt文件储存可变参数
    assert args.config is not None,"you must give your model config"
    # config文件为训练前指定，可选择文件为configs文件夹中的文件
    # 此处必须指定config文件，否则代码无法继续运行
    cfg = Config.fromfile(args.config)		
    # 此处将指定的config文件中的参数传入元祖cfg中
    
    
      
      
      
      
      
      
    

    train_data_set = build_dataset(cfg.data.train)
    # 通过build_dataset()函数生成训练数据集
    train_loader = DataLoader(train_data_set,batch_size=opt.trainBatch,shuffle=True,num_workers=4,collate_fn=train_loader_collate_fn)
    # 采用pytorch内置函数进行训练数据的载入
    test_data_set = build_dataset(cfg.data.test)
    # 通过build_dataset()函数生成测试数据集
    test_loader = DataLoader(test_data_set,batch_size=opt.validBatch,shuffle=False,num_workers=4,collate_fn=test_loader_collate_fn)
    # 采用pytorch内置函数进行测试数据的载入
    
    
      
      
      
      
      
      
      
      
    

    def build_dataset(cfg, default_args=None):
    dataset = build_from_cfg(cfg, DATASETS, default_args)
    # 将cfg参数、dataset传入build_from_cfg()函数中
    return dataset
    
    
      
      
      
      
    

    # 此函数的参数为可选cfg参数和registry类
    def build_from_cfg(cfg, registry, default_args=None):
    """Build a module from config dict.
    	创建一个基于config字典的模型
    Args:
        cfg (dict): Config dict. It should at least contain the key "type".
        registry (:obj:`Registry`): The registry to search the type from.
        default_args (dict, optional): Default initialization arguments.
    
    Returns:
        obj: The constructed object.
    """
    assert isinstance(cfg, dict) and 'type' in cfg
    assert isinstance(default_args, dict) or default_args is None
    args = cfg.copy()
    obj_type = args.pop('type')
    if str(obj_type):
        obj_cls = registry.get(obj_type)
    
        if obj_cls is None:
            raise KeyError('{} is not in the {} registry'.format(
                obj_type, registry.name))
    elif inspect.isclass(obj_type):
        obj_cls = obj_type
    else:
        raise TypeError('type must be a str or valid type, but got {}'.format(
            type(obj_type)))
    if default_args is not None:
        for name, value in default_args.items():
            args.setdefault(name, value)
    return obj_cls(**args)
    
    
      
      
      
      
      
      
      
      
      
      
      
      
      
      
      
      
      
      
      
      
      
      
      
      
      
      
      
      
      
      
      
    

    cfg.checkpoints = os.path.join(cfg.checkpoints,cfg.name)
    if not os.path.exists(cfg.checkpoints):
        os.mkdir(cfg.checkpoints)
    
    
      
      
      
    

    if "Alpha.py" in args.config:
        pose_generator =None
    else:
    # 此处调用build_generator()函数生成姿态
        pose_generator = build_generator(cfg.pose_generator)
    device = torch.device("cuda")	# 调用cuda
    
    
      
      
      
      
      
      
    

    def build_generator(cfg):
    return build(cfg, GENERATOR)
    
    
      
      
    

    def build(cfg, registry, default_args=None):
    if isinstance(cfg, list):
        modules = [
            build_from_cfg(cfg_, registry, default_args) for cfg_ in cfg
        ]
        return nn.Sequential(*modules)
    else:
        return build_from_cfg(cfg, registry, default_args)
        # 此处函数直接返回build_from_cfg()函数生成的值，生成模型
    
    
      
      
      
      
      
      
      
      
      
    

    # 首先调用engineer/models/builder.py中的
    # build_backbone()函数去生成backbone模型
    model_pos = build_backbone(cfg.model).to(device).to(device)
    model_pos.train() # 此处调用pytorch内置的train()函数
    
    
      
      
      
      
    

    def build_backbone(cfg):
    return build(cfg, BACKBONES)
    
    
      
      
    

    def build(cfg, registry, default_args=None):
    if isinstance(cfg, list):
        modules = [
            build_from_cfg(cfg_, registry, default_args) for cfg_ in cfg
        ]
        return nn.Sequential(*modules)
    else:
        return build_from_cfg(cfg, registry, default_args)
    
    
      
      
      
      
      
      
      
      
    

    param_list = []
    for module in model_pos.gcn_head:
    params_list = group_weight.group_weight(param_list,module,args.LR)
    for module in model_pos.heat_map_head:
    params_list = group_weight.group_weight(param_list,module,1e-3)
    for module in model_pos.generator_map:
    params_list = group_weight.group_weight(param_list,module,1e-3)
    criterion = nn.L1Loss(size_average=True,reduce=True).to(device)
    optimizer = torch.optim.Adam(params_list, lr=args.LR)
    cfg.LR=args.LR
    cfg.nEpochs = args.nEpochs
    
    
      
      
      
      
      
      
      
      
      
      
      
    

    train_epochs(model_pos, optimizer, cfg, args, train_loader, pose_generator, criterion,test_loader,cfg.pred_json)
    
    
      
    

    // An highlighted block
    var foo = 'bar';
    
    
      
      
    

 Luke