Custom Mapper tutorial does not work

📚 Documentation Issue or not clear

Hello, I am following this in order to replycate how to use a custom mapper.

So if I just copy the example in the documentation

def custom_mapper(dataset_dict):
    dataset_dict = copy.deepcopy(dataset_dict)  # it will be modified by code below
    # can use other ways to read image
    image = utils.read_image(dataset_dict["file_name"], format="BGR")
    # See "Data Augmentation" tutorial for details usage
    auginput = T.AugInput(image)
    transform = T.Resize((800, 800))(auginput)
    image = torch.from_numpy(auginput.image.transpose(2, 0, 1))
    annos = [
        utils.transform_instance_annotations(annotation, [transform], image.shape[1:])
        for annotation in dataset_dict.pop("annotations")
    ]
    return {
        # create the format that the model expects
        "image": image,
        "instances": utils.annotations_to_instances(annos, image.shape[1:])
    }

And then, follwing this i overwrited the build_{train,test}_loader method

class MyTrainer(DefaultTrainer):
    @classmethod
    def build_train_loader(cls, cfg):
        train_dataloader =  build_detection_train_loader(cfg, mapper=custom_mapper)
        return train_dataloader

    @classmethod
    def build_test_loader(cls, cfg):
        test_dataloader = build_detection_train_loader(cfg, mapper=custom_mapper)
        return test_dataloader

I get this error when start to train

w, h = d["width"], d["height"]
KeyError: 'width'

[mjack3]

@narenakash

Yes, if you check the documentation, it says

# create the format that the model expects In my case, it is faster 50, so the input model format acording to the documentation also needs:

“height”, “width”:

[narenakash]

@mjack3 I’m also facing the same issue. Could you share how you solved it?

Thanks in advance.