Inference with Visual Segmentation model without label

[yit.chan]

I have my model trained with visual segmentation as pretrain model. As per visual segmentation, it seems requires A- new input image, B-golden reference image and label image during training.
Since it is a new input image, I am manage to create a golden reference image to the new image, while pretending that it has no any expectation on what the differences, i will not have any label image (mask image). How can I ran the inference to test my new input image without referencing to the label image. I had tried to leave it empty annotation_folder_name: “” but it is still looking for the label image when i start inference it.

multi_scale_train: True
multi_scale_infer: False
num_classes: 2
img_size: 816
image_folder_name: “A”
change_image_folder_name: ‘B’
list_folder_name: ‘list’
annotation_folder_name: “”
train_split: “train-1”
validation_split: “val-1”
test_split: ‘test-good’
predict_split: ‘test-good’

[yit.chan]

It looks like it has to manually create a label image referencing to the input image. Even a full black or white image should be fine. Now I manage to get the inference run now.