DGS Types

[SerdarHelli]

On DGS types, there are multi glosses corresponding one pose file. Is it possible ?

For example :

glosses=[ gloss1,gloss2,gloss3]

corresponding to one pose file according to id ?

shoul I take gloss1 ?

I am trying to create dictionary like :

dict={ gloss : correspondig pose ,
....}

[AmitMY]

It is possible, and you should probably take all of them.
This happens when the sign has the same form but multiple meanings.

Looking at types you can see it:

In this screenshot, you see OBERFLÄCHE1 and BEREICH1A^ are both the same form https://www.sign-lang.uni-hamburg.de/meinedgs/types/type13215_de.html

[SerdarHelli]

Also , I am gettin this error ,

config = SignDatasetConfig(name="dgs_types_poses", version="1.0.0", include_video=False, process_video=False, include_pose="holistic")
dgs_types = tfds.load('dgs_types', builder_kwargs=dict(config=config), data_dir='/content/drive/MyDrive/sing_language_datasets')

TypeError: Error while serializing feature views/pose/data: TensorInfo(shape=(None, None, 1, 576, 3), dtype=float32): 'NoneType' object cannot be interpreted as an integer

[SerdarHelli]

It is possible, and you should probably take all of them. This happens when the sign has the same form but multiple meanings.

Looking at types you can see it:

In this screenshot, you see OBERFLÄCHE1 and BEREICH1A^ are both the same form https://www.sign-lang.uni-hamburg.de/meinedgs/types/type13215_de.html

But in some samples ,the glosses are quite different each other. Should I accept these have same pose ?

[SerdarHelli]

I am doing like this

config = SignDatasetConfig(name="dgs_types_video", version="1.0.0", include_video=False, process_video=False, include_pose=None)
dgs_types = tfds.load('dgs_types', builder_kwargs=dict(config=config), data_dir='/content/drive/MyDrive/SignLanguage/dataset_slp')
decode_str = lambda s: s.numpy().decode('utf-8')
c_dict=defaultdict()
for datum in tqdm(dgs_types["train"]):
    _id = decode_str(datum['id'])
    pose_file_path=f"/content/drive/MyDrive/SignLanguage/dgs_poses/{_id}_frontal.pose"
    !wget -q {pose_download_path} -P /content/drive/MyDrive/SignLanguage/dgs_poses
    pose_file_path=f"/content/drive/MyDrive/SignLanguage/dgs_poses/{_id}_frontal.pose"
    if os.path.exists(pose_file_path):
      for gloss in datum["glosses"]:
          gloss=correct_gloss(gloss)
          c_dict[gloss]=f"{_id}_frontal.pose"

[AmitMY]

TypeError: Error while serializing feature views/pose/data: TensorInfo(shape=(None, None, 1, 576, 3), dtype=float32): 'NoneType' object cannot be interpreted as an integer

Could you please open a different issue forthis?

But in some samples ,the glosses are quite different each other. Should I accept these have same pose ?

This is according to the Hamburg people. I think you should accept the same pose despite the different words.

I am doing like this

Your code looks good to me

[SerdarHelli]

Thanks AmitMy . Yours works are amazing :)