def main_worker()

in scripts/action_recognition/test.py [0:0]

• 19 lines of code
• 8 McCabe index (conditional complexity)

def main_worker(cfg):
    cfg.defrost()
    if cfg.CONFIG.TEST.BATCH_SIZE != 1:
        # batch_size during multiview test must be set to 1 due to limited GPU memory
        cfg.CONFIG.TEST.BATCH_SIZE = 1
    if not cfg.CONFIG.TEST.MULTI_VIEW_TEST:
        # enable multi-view testing, e.g., 30-view
        cfg.CONFIG.TEST.MULTI_VIEW_TEST = True
    cfg.freeze()

    # create model
    print('Creating model: %s' % (cfg.CONFIG.MODEL.NAME))
    model = get_model(cfg)
    optimizer = None
    model, _, _ = deploy_model(model, optimizer, cfg)
    num_parameters = sum(p.numel() for p in model.parameters() if p.requires_grad)
    print('Number of parameters in the model: %6.2fM' % (num_parameters / 1000000))
    if cfg.CONFIG.MODEL.LOAD or cfg.CONFIG.MODEL.PRETRAINED:
        model, _ = load_model(model, cfg, load_fc=cfg.CONFIG.MODEL.LOAD_FC)

    # create dataset and dataloader
    train_loader, test_loader, train_sampler, test_sampler, _ = build_dataloader(cfg)

    # create criterion
    criterion = torch.nn.CrossEntropyLoss().cuda()

    # run evaluation
    acc1, acc5, _ = test_classification(cfg, test_loader, model, criterion, 0, None)
    print('Top-1 accuracy: {top1_acc:.2f}%, Top-5 accuracy: {top5_acc:.2f}%'.format(top1_acc=acc1, top5_acc=acc5))