Function test

lib/core_function.py:278–310

(cfg, model, critetion, epoch, data_loader, logger, writer, devices, valIters, metrics_base='combine')

def test(cfg, model, critetion, epoch, data_loader, logger, writer, devices, valIters, metrics_base='combine'):
    calculate_acc = get_acc_mesure_func(metrics_base)
    total_preds = torch.tensor([]).cuda().to(dtype=torch.float64)
    total_labels = torch.tensor([]).cuda().to(dtype=torch.float64)
    
    #Switch to test mode
    model.eval()
    test_dataloader = tqdm(data_loader, dynamic_ncols=True)
    with torch.no_grad():
        for b, (inputs, labels, vid_ids) in enumerate(test_dataloader):
            inputs = inputs.to(dtype=torch.float64).cuda()
            labels = labels.to(dtype=torch.float64).cuda()
            
            outputs = model(inputs)
            # Applying Flip test
            if isinstance(outputs, list):
                outputs = outputs[0]
            
            #In case outputs contain a dict key
            if isinstance(outputs, dict):
                hm_outputs = outputs['hm']
                cls_outputs = outputs['cls']

            total_preds = torch.cat((total_preds, cls_outputs), 0)
            total_labels = torch.cat((total_labels, labels), 0)
            
        acc_ = calculate_acc(total_preds, targets=None, labels=total_labels, threshold=cfg.TEST.threshold)
        auc_, ap_, ar_, mf1_ = bin_calculate_auc_ap_ar(total_preds, total_labels, metrics_base=metrics_base)
        
        logger.info(f'Current ACC, AUC, AP, AR, mF1 for {cfg.DATASET.DATA.TEST.FAKETYPE} --- {cfg.DATASET.DATA.TEST.LABEL_FOLDER} -- \
            {acc_*100} -- {auc_*100} -- {ap_*100} -- {ar_*100} -- {mf1_*100}')

    return acc_, auc_, ap_, ar_