Datasets:

yinliang22
/

matcurvs

Exception:    SplitsNotFoundError
Message:      The split names could not be parsed from the dataset config.
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 286, in get_dataset_config_info
                  for split_generator in builder._split_generators(
                                         ~~~~~~~~~~~~~~~~~~~~~~~~~^
                      StreamingDownloadManager(base_path=builder.base_path, download_config=download_config)
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/folder_based_builder/folder_based_builder.py", line 250, in _split_generators
                  raise ValueError(
                      "`file_name`, `*_file_name`, `file_names` or `*_file_names` must be present as dictionary key in metadata files"
                  )
              ValueError: `file_name`, `*_file_name`, `file_names` or `*_file_names` must be present as dictionary key in metadata files
              
              The above exception was the direct cause of the following exception:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/split_names.py", line 68, in compute_split_names_from_streaming_response
                  for split in get_dataset_split_names(
                               ~~~~~~~~~~~~~~~~~~~~~~~^
                      path=dataset,
                      ^^^^^^^^^^^^^
                      config_name=config,
                      ^^^^^^^^^^^^^^^^^^^
                      token=hf_token,
                      ^^^^^^^^^^^^^^^
                  )
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 340, in get_dataset_split_names
                  info = get_dataset_config_info(
                      path,
                  ...<6 lines>...
                      **config_kwargs,
                  )
                File "/usr/local/lib/python3.14/site-packages/datasets/inspect.py", line 291, in get_dataset_config_info
                  raise SplitsNotFoundError("The split names could not be parsed from the dataset config.") from err
              datasets.inspect.SplitsNotFoundError: The split names could not be parsed from the dataset config.

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

MatCurvs

The benchmark companion to MatDeplot — an article-scale, real-coordinate curve-extraction system for materials-science line plots (NeurIPS 2026 Datasets and Benchmarks track).

Code: https://github.com/yinliang420/matcurvs

Three subsets, one progression

Layer	Subset	Size	Annotation depth	Use
L0	`l0/`	~1.4M subfigures	chart-type tag + 3×10 taxonomy	subfigure routing — find the line plots in 55,763 articles
L1	`l1/`	204 panels	per-curve pixel polyline + axis-calibrated (x, y)	pixel-anchor evaluation (LR_θ); 90-panel leakage-clean hold
L2	`l2/`	3,817 panels / 8,026 curves	verified real-coordinate (x, y) sequences + reasoning Q-bank	agentic numerical-reasoning evaluation

Headline numbers (NeurIPS 2026 D&B paper)

Failure mode	Best hosted VLM	MatDeplot	Gap
Pixel anchoring (LR_0.5 on n=90 leakage-clean L1)	≤1.2%	45.8%	38×
`y` at x_0 on 14,740 Q (image only)	99.8% rel-err	—	—
Same task, image-blind LLM with extracted (x,y)	—	1.5%	67×
EIS Z′ intercept (image only)	100% rel-err	—	—
Same task, with (x,y) data	—	0%	∞

Construction

L0 → L1 → L2 trace the same panels through three escalating annotation regimes:

L0 (subfigure routing). Every figure of every article in a 55,763-paper open-access materials-science corpus is broken into panels by a YOLOv11x detector and tagged by a multimodal-LLM classifier (1,375,165 panels, 47.81% classified as line charts).
L1 (pixel polylines). 204 line-chart panels are hand-traced by a human annotator: every curve is a polyline in pixel space, then re-projected to the chart's calibrated real-coordinate axes. A leakage-clean n=90 hold is set aside for the headline pixel-IoU metric.
L2 (real-coordinate curves). The ~657K line-chart panels are processed by the MatDeplot extractor (Mask2Former segmenter + 5D LAB-xy k-means + axis calibration) and curated by a multimodal-LLM judge against curve count, shape, axis-range plausibility, and artifact-freeness. The 3,817 panels that pass every check, with their 8,026 real-coordinate curves, are released here. The l2/matcurvs_reasoning/ directory bundles a 14,740-question numerical-reasoning task built deterministically over those curves.

Download

pip install huggingface_hub
# all three subsets
huggingface-cli download yinliang22/matcurvs --repo-type dataset --local-dir ./matcurvs

# or just one tier
huggingface-cli download yinliang22/matcurvs --repo-type dataset \
    --include "l1/*" --local-dir ./matcurvs

Citation

To appear at NeurIPS 2026 (Datasets and Benchmarks track). Citation information will be added on acceptance.

License

CC-BY-4.0. The annotations are released under CC-BY-4.0; the underlying chart panels remain under the licence of the source open-access publication.

Downloads last month: 2

Total file size:

1.56 GB