Problems when saving the predictions of ensemble evaluator

[cugwu]

Hi, I'm using monai.engines.EnsembleEvaluator. I want to save my final predictions after the ensemble. Following the tutorial I have:

def ensemble_evaluate(post_transforms, models, pred_keys, device, loader, inferer, metrics_saver):
    evaluator = EnsembleEvaluator(
        device=device,
        val_data_loader=loader,
        pred_keys=pred_keys,
        networks=models,
        inferer=inferer,
        postprocessing=post_transforms,
        val_handlers=metrics_saver,
        key_val_metric={
            "test_mean_dice": MeanDice(
                include_background=True,
                reduction='mean',
                output_transform=from_engine(["pred", "label"])
            ),
    )
    evaluator.run()

....

mean_post_transforms = Compose(
            [
                EnsureTyped(keys=pred_keys),
                MeanEnsembled(
                    keys=pred_keys,
                    output_key="pred",
                    weights=w_mee,
                ),
                Activationsd(keys="pred", softmax=True),
                AsDiscreted(keys="pred", argmax=True, to_onehot=args.out_channels),
                AsDiscreted(keys="label", to_onehot=args.out_channels),
                SaveImaged(keys="pred", output_dir=output_directory, output_postfix="mean",
                           output_ext=".nii.gz", resample=False)
            ]
                  
 ensemble_evaluate(mean_post_transforms, models, pred_keys, torch.device(device), test_loader,
                          test_inferer, metrics_saver)

I know that SaveImaged should save your final volume but when I try to open it with the following code:

train_images = sorted(glob.glob(os.path.join(data_dir, "img", "img0062.nii.gz")))
    train_labels = sorted(glob.glob(os.path.join(data_dir, "pred", "img0062.nii.gz")))
    data_dicts = [
        {"image": image_name, "label": label_name}
        for image_name, label_name in zip(train_images, train_labels)
    ]
    transform = Compose([
        LoadImaged(keys=["image", "label"]),
        EnsureChannelFirstd(keys=["image", "label"]),
        Orientationd(keys=["image", "label"], axcodes="PLS"),
        Spacingd(keys=["image", "label"], pixdim=(1.5, 1.5, 2.0), mode=("bilinear", "nearest")),
        ScaleIntensityRanged(keys=["image"], a_min=-57, a_max=164, b_min=0.0, b_max=1.0, clip=True),
        CropForegroundd(keys=["image", "label"], source_key="image"),
    ])

    check_ds = Dataset(data=data_dicts, transform=transform)
    check_loader = DataLoader(check_ds, batch_size=1)
    data = first(check_loader)
    print(f"image shape: {data['image'].shape}, label shape: {data['label'].shape}")

I have the following error:
File "/home/clusterusers/cugwu/codes/wacv2023/BTCV/visulize_results.py", line 51, in <module> print(f"image shape: {data['image'].shape}, label shape: {data['label'].shape}") TypeError: 'NoneType' object is not subscriptable

What I'm doing wrong? Why I'm not able to save the predictions of the ensemble evaluator and visualize them? I have also tried to put in SaveImaged the following parameter: meta_keys="image_meta_dict" but nothing changed.

[KumoLiu]

Hi @cugwu, could you please check your data type, since your error message showed that it is None.

[cugwu]

yes, there was a problem with the data type I don't know why but it was wrongly saved. Thank you for the answer

[KumoLiu]

Hi @cugwu, try testing with a small piece of code to find the error further, if you have any further questions, feel free to ask me, thanks!