Datasets:
Tasks:
Image Classification
Sub-tasks:
multi-class-image-classification
Languages:
English
Size:
100K<n<1M
ArXiv:
License:
Help required to make data possible to load by others
#2
by
jordyvl
- opened
@lhoestq
This dataset has quite some value for the document AI community, but TBH the setup is driving me nuts.
There seem to be issues with loading npz files, as well as issues with parsing DatasetInfo.
The difference between _info and info and where actual features come from/will be defined is severely underdocumented and makes it very hard to just make an easy data loading script.
Could you have a look at where I have messed it up?
Also default config name does not always seem to trigger...
This check does pass:datasets-cli test rvl_cdip_easyOCR.py --save_info --all_configs
(mp_docvqa) jordy@frodo:~/code/Beyond-Document-Page-Classification/src/data/rvl-cdip_easyOCR$ datasets-cli test rvl_cdip_easyOCR.py --save_info --all_configs
Testing builder 'default' (1/1)
Downloading and preparing dataset rvl_cdip_easy_ocr/default (download: 36.12 GiB, generated: 45.24 GiB, post-processed: Unknown size, total: 81.35 GiB) to /home/jordy/.cache/huggingface/datasets/rvl_cdip_easy_ocr/default/1.0.0/45af7ad8b9ae9e8946d48ba8882e3f5d2e101a7a4adb195aa86d1aff87d65f66...
Computing checksums of downloaded files. They can be used for integrity verification. You can disable this by passing ignore_verifications=True to load_dataset
Computing checksums: 100%|███████████████████████████████████████████████████████████████████████████████████████████| 1/1 [05:26<00:00, 326.45s/it]
Downloading data files: 100%|███████████████████████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 1649.14it/s]
Loading OCR data: 100%|██████████████████████████████████████████████████████████████████████████████████| 319837/319837 [00:17<00:00, 18180.12it/s]
Loading OCR data: 100%|████████████████████████████████████████████████████████████████████████████████████| 39996/39996 [00:01<00:00, 22571.31it/s]
Loading OCR data: 100%|████████████████████████████████████████████████████████████████████████████████████| 39995/39995 [00:02<00:00, 18352.99it/s]
Dataset rvl_cdip_easy_ocr downloaded and prepared to /home/jordy/.cache/huggingface/datasets/rvl_cdip_easy_ocr/default/1.0.0/45af7ad8b9ae9e8946d48ba8882e3f5d2e101a7a4adb195aa86d1aff87d65f66. Subsequent calls will reuse this data.
100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 3/3 [00:08<00:00, 2.92s/it]
Dataset card saved at README.md
Test successful.
When loading the dataset, it triggers this error:
Traceback (most recent calls WITHOUT Sacred internals):
File "/home/jordy/miniconda3/envs/mp_docvqa/lib/python3.9/site-packages/datasets/builder.py", line 1833, in _prepare_split_single
writer.write_table(table)
File "/home/jordy/miniconda3/envs/mp_docvqa/lib/python3.9/site-packages/datasets/arrow_writer.py", line 567, in write_table
pa_table = table_cast(pa_table, self._schema)
File "/home/jordy/miniconda3/envs/mp_docvqa/lib/python3.9/site-packages/datasets/table.py", line 2312, in table_cast
return cast_table_to_schema(table, schema)
File "/home/jordy/miniconda3/envs/mp_docvqa/lib/python3.9/site-packages/datasets/table.py", line 2270, in cast_table_to_schema
raise ValueError(f"Couldn't cast\n{table.schema}\nto\n{features}\nbecause column names don't match")
ValueError: Couldn't cast
text: string
to
{'id': Value(dtype='string', id=None), 'image': Image(decode=True, id=None), 'label': ClassLabel(names=['letter', 'form', 'email', 'handwritten', 'advertisement', 'scientific report', 'scientific publication', 'specification', 'file folder', 'news article', 'budget', 'invoice', 'presentation', 'questionnaire', 'resume', 'memo'], id=None), 'words': Sequence(feature=Value(dtype='string', id=None), length=-1, id=None), 'boxes': Sequence(feature=Sequence(feature=Value(dtype='int32', id=None), length=-1, id=None), length=-1, id=None)}
because column names don't match
This seems to load an empty, non-existing config...
Even when explicitly passing a config:
>>> load_dataset("jordyvl/rvl-cdip_easyOCR",'default')
Using custom data configuration default-3da9bc025fd0c829
Downloading and preparing dataset None/None to /home/jordy/.cache/huggingface/datasets/jordyvl___text/default-3da9bc025fd0c829/0.0.0/cb1e9bd71a82ad27976be3b12b407850fe2837d80c22c5e03a28949843a8ace2...
I don't understand what you're trying to achieve in your dataset script. You should:
- download the data files in _split_generators using the dl_manager and define your splits
- open the data files in _generate_examples and yield examples
You can pass URLs or relative paths to data files to the dl_manager
you can debug locally using
load_dataset("path/to/local/rvl_cdip_easyOCR")