How to download the dataset?

#1
by YandanYang - opened

I tried load_dataset("hssd/hssd-hab") and load_dataset("hssd-hab").
It says " Couldn't find 'hssd/hssd-hab' on the Hugging Face Hub either: FileNotFoundError: Dataset 'hssd/hssd-hab' doesn't exist on the Hub. If the repo is private or gated, make sure to log in with huggingface-cli login."
Also, downloading it with git failed and stuck in 95% process.

Habitat Synthetic Scenes Dataset org

Hello @YandanYang ,

Currently, we don’t support downloading from HuggingFace's Python API, but git cloning the repository should work.

I am not sure why your git clone failed. You should be running the following commands (after accepting the conditions to access the dataset on the repository webpage).

git lfs install
git clone https://huggingface.co/datasets/hssd/hssd-hab

Ok. Thanks.
I guess it is the problem of my pc, since git works well on my laptop.

I have another question.
I do not find any information about the category in both the object and scene folder.
Do you release the category information? Where can I find it?
By the way, what is the meaning of foleder named "0-x" in the folder of objects? Why are the assets splited into these folders?

Thanks.

Habitat Synthetic Scenes Dataset org

Do you release the category information? Where can I find it?

We have added annotations for objects belonging to 28 common object categories through the semantic_id attribute in the object.config.json files (example). We plan to include the semantic-id-to-category mapping in an upcoming release (in a few days). Until then, here is the mapping:

{
    "alarm_clock": 1,
    "bed": 2,
    "book": 3,
    "bottle": 4,
    "bowl": 5,
    "chair": 6,
    "chest_of_drawers": 7,
    "couch": 8,
    "cushion": 9,
    "drinkware": 10,
    "fridge": 11,
    "laptop": 12,
    "microwave": 13,
    "picture": 14,
    "plate": 15,
    "potted_plant": 16,
    "shelves": 17,
    "shoes": 18,
    "sink": 19,
    "stool": 20,
    "table": 21,
    "table_lamp": 22,
    "toaster": 23,
    "toilet": 24,
    "trashcan": 25,
    "tv": 26,
    "vase": 27,
    "washer_dryer": 28
}

Objects that do not belong to any of these 28 categories do not have the semantic_id specified in their config files. To get semantic information about all the 18k+ objects (along with additional metadata), please refer to this CSV file. We plan to include this in the upcoming release too.

By the way, what is the meaning of foleder named "0-x" in the folder of objects? Why are the assets splited into these folders?

The object assets were split across multiple folders because HuggingFace only allows uploading only a limited number of files in each folder.

Ok. That helps a lot~ Thanks.
Since there will be more release info , I want to know will there be more details about the room types in each scene?
Now the objects are listed together in a whole scene. I do not know which room each object is belongs to. Will there be this kind of information?

Hi,
To get semantic information about all the 18k+ objects (along with additional metadata), please refer to this CSV file.
I have check the CSV file. Will you release the articulated information, such as urdf file including joints and links, of the 1k+ articulatable objects?

Habitat Synthetic Scenes Dataset org

Hello @YandanYang , we currently do not have the information about the rooms and articulated objects that you are requesting.

mukulkhanna changed discussion status to closed

Sign up or log in to comment