Dataset Viewer
Auto-converted to Parquet Duplicate
userId
int64
sampleId1
int64
sampleId2
int64
referenceSentenceId1
int64
referenceSentenceId2
int64
questionId
int64
score
int64
submissionTimestamp
large_string
samplePath1
large_string
samplePath2
large_string
3
270,155
270,871
1
3
1
-1
2025-08-22T13:10:51
1/270155.npy
3/270871.npy
3
270,155
271,579
1
9
1
1
2025-08-25T08:31:06
1/270155.npy
9/271579.npy
3
270,155
274,176
1
2
1
-1
2025-08-21T11:03:41
1/270155.npy
2/274176.npy
3
270,155
274,323
1
4
1
1
2025-07-29T12:32:22
1/270155.npy
4/274323.npy
3
270,155
276,032
1
8
1
1
2025-08-11T10:26:19
1/270155.npy
8/276032.npy
3
270,155
277,999
1
1
1
-1
2025-08-24T10:58:22
1/270155.npy
1/277999.npy
3
270,155
278,452
1
1
1
-1
2025-08-23T11:22:39
1/270155.npy
1/278452.npy
3
270,155
279,416
1
8
1
1
2025-08-25T06:57:53
1/270155.npy
8/279416.npy
3
270,155
279,470
1
6
1
1
2025-08-25T12:10:38
1/270155.npy
6/279470.npy
3
270,155
280,162
1
9
1
1
2025-08-14T20:43:38
1/270155.npy
9/280162.npy
3
270,155
280,877
1
4
1
-1
2025-08-06T14:56:22
1/270155.npy
4/280877.npy
3
270,155
280,924
1
1
1
-1
2025-08-14T21:51:03
1/270155.npy
1/280924.npy
3
270,155
281,746
1
4
1
1
2025-07-28T12:49:01
1/270155.npy
4/281746.npy
3
270,156
271,303
2
1
1
1
2025-08-25T08:35:28
2/270156.npy
1/271303.npy
3
270,156
272,392
2
5
1
1
2025-08-24T10:34:28
2/270156.npy
5/272392.npy
3
270,156
272,720
2
5
1
-1
2025-08-16T15:13:19
2/270156.npy
5/272720.npy
3
270,156
274,463
2
8
1
-1
2025-08-23T11:29:45
2/270156.npy
8/274463.npy
3
270,156
275,855
2
3
1
-1
2025-08-25T08:49:24
2/270156.npy
3/275855.npy
3
270,156
276,031
2
7
1
1
2025-08-12T18:47:47
2/270156.npy
7/276031.npy
3
270,156
276,916
2
8
1
-1
2025-08-22T14:05:09
2/270156.npy
8/276916.npy
3
270,156
277,351
2
6
1
1
2025-08-25T10:52:23
2/270156.npy
6/277351.npy
3
270,156
277,497
2
5
1
-1
2025-08-22T14:06:53
2/270156.npy
5/277497.npy
3
270,156
278,458
2
7
1
-1
2025-08-20T22:57:06
2/270156.npy
7/278458.npy
3
270,156
279,075
2
1
1
-1
2025-08-11T10:21:50
2/270156.npy
1/279075.npy
3
270,156
279,415
2
7
1
-1
2025-08-10T15:17:33
2/270156.npy
7/279415.npy
3
270,156
280,286
2
4
1
1
2025-08-16T15:47:01
2/270156.npy
4/280286.npy
3
270,156
281,883
2
2
1
-1
2025-08-11T09:57:17
2/270156.npy
2/281883.npy
3
270,156
281,989
2
8
1
-1
2025-08-25T10:50:45
2/270156.npy
8/281989.npy
3
270,157
270,280
3
9
1
-1
2025-07-28T12:38:11
3/270157.npy
9/270280.npy
3
270,157
270,284
3
4
1
1
2025-08-25T09:10:57
3/270157.npy
4/270284.npy
3
270,157
270,589
3
8
1
1
2025-08-25T10:43:13
3/270157.npy
8/270589.npy
3
270,157
270,870
3
2
1
-1
2025-08-20T23:21:22
3/270157.npy
2/270870.npy
3
270,157
270,875
3
7
1
-1
2025-08-23T11:43:22
3/270157.npy
7/270875.npy
3
270,157
273,620
3
6
1
-1
2025-08-22T13:50:24
3/270157.npy
6/273620.npy
3
270,157
280,070
3
8
1
1
2025-08-25T10:47:15
3/270157.npy
8/280070.npy
3
270,157
280,291
3
9
1
1
2025-08-14T21:30:44
3/270157.npy
9/280291.npy
3
270,157
280,435
3
3
1
-1
2025-08-25T11:56:58
3/270157.npy
3/280435.npy
3
270,157
280,445
3
4
1
-1
2025-08-20T22:34:26
3/270157.npy
4/280445.npy
3
270,157
280,938
3
6
1
-1
2025-08-14T21:51:10
3/270157.npy
6/280938.npy
3
270,157
281,072
3
4
1
-1
2025-08-07T10:12:43
3/270157.npy
4/281072.npy
3
270,157
281,751
3
9
1
1
2025-08-21T12:11:25
3/270157.npy
9/281751.npy
3
270,157
281,842
3
1
1
1
2025-08-20T23:32:31
3/270157.npy
1/281842.npy
3
270,157
281,940
3
9
1
-1
2025-08-11T11:11:18
3/270157.npy
9/281940.npy
3
270,158
270,276
4
5
1
-1
2025-08-12T18:25:37
4/270158.npy
5/270276.npy
3
270,158
270,363
4
3
1
-1
2025-08-17T14:23:39
4/270158.npy
3/270363.npy
3
270,158
271,205
4
4
1
-1
2025-08-25T10:30:57
4/270158.npy
4/271205.npy
3
270,158
272,925
4
3
1
-1
2025-08-22T13:20:59
4/270158.npy
3/272925.npy
3
270,158
274,013
4
5
1
-1
2025-08-22T12:09:04
4/270158.npy
5/274013.npy
3
270,158
274,326
4
7
1
-1
2025-08-07T10:17:01
4/270158.npy
7/274326.npy
3
270,158
274,407
4
1
1
-1
2025-08-21T11:50:44
4/270158.npy
1/274407.npy
3
270,158
277,294
4
1
1
-1
2025-08-25T09:01:06
4/270158.npy
1/277294.npy
3
270,158
277,302
4
9
1
1
2025-08-11T10:14:08
4/270158.npy
9/277302.npy
3
270,158
278,008
4
10
1
-1
2025-08-13T19:37:56
4/270158.npy
10/278008.npy
3
270,158
278,279
4
4
1
1
2025-08-11T11:12:31
4/270158.npy
4/278279.npy
3
270,158
278,312
4
9
1
1
2025-08-25T08:31:45
4/270158.npy
9/278312.npy
3
270,158
278,982
4
4
1
1
2025-08-11T09:54:44
4/270158.npy
4/278982.npy
3
270,158
279,096
4
5
1
-1
2025-08-22T13:01:34
4/270158.npy
5/279096.npy
3
270,158
280,154
4
1
1
-1
2025-08-11T10:12:34
4/270158.npy
1/280154.npy
3
270,158
281,069
4
1
1
-1
2025-08-17T13:42:37
4/270158.npy
1/281069.npy
3
270,158
281,465
4
1
1
-1
2025-08-25T12:12:05
4/270158.npy
1/281465.npy
3
270,159
270,590
5
9
1
1
2025-07-29T12:08:38
5/270159.npy
9/270590.npy
3
270,159
271,428
5
1
1
-1
2025-08-22T14:26:36
5/270159.npy
1/271428.npy
3
270,159
271,429
5
2
1
-1
2025-08-25T08:34:25
5/270159.npy
2/271429.npy
3
270,159
272,156
5
2
1
1
2025-08-21T00:14:29
5/270159.npy
2/272156.npy
3
270,159
272,924
5
2
1
-1
2025-08-07T10:10:45
5/270159.npy
2/272924.npy
3
270,159
272,996
5
7
1
-1
2025-08-25T12:25:17
5/270159.npy
7/272996.npy
3
270,159
276,483
5
3
1
-1
2025-08-25T10:43:08
5/270159.npy
3/276483.npy
3
270,159
278,285
5
10
1
1
2025-08-25T10:39:54
5/270159.npy
10/278285.npy
3
270,159
278,308
5
5
1
-1
2025-08-25T10:39:17
5/270159.npy
5/278308.npy
3
270,159
279,078
5
4
1
1
2025-08-13T19:50:43
5/270159.npy
4/279078.npy
3
270,159
279,469
5
5
1
1
2025-07-28T12:43:45
5/270159.npy
5/279469.npy
3
270,159
279,931
5
3
1
-1
2025-08-11T09:56:14
5/270159.npy
3/279931.npy
3
270,159
280,695
5
3
1
-1
2025-07-28T12:41:02
5/270159.npy
3/280695.npy
3
270,159
281,077
5
9
1
1
2025-08-22T14:22:24
5/270159.npy
9/281077.npy
3
270,159
281,584
5
9
1
-1
2025-07-29T12:33:27
5/270159.npy
9/281584.npy
3
270,160
271,309
6
7
1
1
2025-08-25T08:34:04
6/270160.npy
7/271309.npy
3
270,160
271,761
6
1
1
-1
2025-08-10T16:38:22
6/270160.npy
1/271761.npy
3
270,160
272,258
6
2
1
1
2025-07-28T13:05:27
6/270160.npy
2/272258.npy
3
270,160
273,258
6
5
1
-1
2025-08-25T10:48:13
6/270160.npy
5/273258.npy
3
270,160
274,016
6
8
1
-1
2025-08-25T08:49:11
6/270160.npy
8/274016.npy
3
270,160
274,179
6
5
1
-1
2025-08-15T19:06:24
6/270160.npy
5/274179.npy
3
270,160
277,755
6
9
1
-1
2025-08-03T15:28:23
6/270160.npy
9/277755.npy
3
270,160
278,087
6
2
1
-1
2025-08-22T14:04:44
6/270160.npy
2/278087.npy
3
270,160
279,419
6
2
1
-1
2025-08-24T10:19:45
6/270160.npy
2/279419.npy
3
270,160
279,426
6
9
1
-1
2025-08-07T10:23:14
6/270160.npy
9/279426.npy
3
270,160
279,436
6
1
1
1
2025-08-11T10:29:49
6/270160.npy
1/279436.npy
3
270,160
280,286
6
4
1
1
2025-08-12T19:08:45
6/270160.npy
4/280286.npy
3
270,161
270,704
7
9
1
1
2025-08-14T21:06:30
7/270161.npy
9/270704.npy
3
270,161
271,770
7
10
1
-1
2025-08-22T12:30:24
7/270161.npy
10/271770.npy
3
270,161
272,511
7
5
1
-1
2025-08-25T08:19:41
7/270161.npy
5/272511.npy
3
270,161
273,621
7
7
1
-1
2025-07-28T12:29:01
7/270161.npy
7/273621.npy
3
270,161
273,647
7
5
1
-1
2025-08-12T18:47:24
7/270161.npy
5/273647.npy
3
270,161
276,483
7
3
1
-1
2025-08-20T23:31:33
7/270161.npy
3/276483.npy
3
270,161
276,714
7
2
1
1
2025-08-15T18:58:44
7/270161.npy
2/276714.npy
3
270,161
278,079
7
3
1
-1
2025-08-24T11:32:18
7/270161.npy
3/278079.npy
3
270,161
278,080
7
4
1
-1
2025-07-23T15:55
7/270161.npy
4/278080.npy
3
270,161
278,311
7
8
1
-1
2025-08-14T21:22:51
7/270161.npy
8/278311.npy
3
270,161
279,076
7
3
1
-1
2025-08-21T11:30:39
7/270161.npy
3/279076.npy
3
270,161
280,000
7
5
1
-1
2025-08-20T23:11:13
7/270161.npy
5/280000.npy
3
270,162
271,763
8
3
1
-1
2025-07-23T15:32
8/270162.npy
3/271763.npy
End of preview. Expand in Data Studio

HaLO: Handwriting Assessment for Legibility and Ordering

HaLO is a dataset of 1,907 handwriting images from 202 German schoolchildren (ages 10--12), annotated with 19,070 pairwise legibility judgments. It supports research on AI-based handwriting legibility assessment through comparative ranking.

Dataset Description

Each handwriting sample depicts one of ten predefined German sentences (20--27 characters each), written by a child in their natural handwriting style. The samples were scanned and cropped from paper sheets. Due to privacy constraints, only Pixtral 12B image embeddings (1024-dimensional feature vectors) are released, not the raw images.

Annotation

Legibility annotations were collected using a pairwise comparison paradigm: annotators were shown two handwriting samples side by side and asked to select the more legible one. A pre-study compared this approach with absolute Likert-scale ratings and found that pairwise comparisons produce higher inter- and intra-rater agreement. The main study annotations were collected by eight annotators (three experts, five non-experts).

Data Splits

The dataset is split into training (1,134 samples, 121 children), validation (382 samples, 40 children), and test (391 samples, 41 children) partitions. All samples from the same child are assigned to the same partition to prevent information leakage.

Configurations

default / main_study

Pairwise legibility annotations from the main study. Each row is a comparison between two samples.

Field Description
sampleId1, sampleId2 IDs of the two compared samples
score 1 if sample 1 is more legible, -1 if sample 2 is more legible
samplePath1, samplePath2 Paths to the corresponding feature .npy files
userId Annotator ID
referenceSentenceId1, referenceSentenceId2 Reference sentence IDs (1--10)
questionId Annotation question ID
submissionTimestamp Timestamp of the annotation

Splits: train (11,340), test (3,910), validation (3,820)

from datasets import load_dataset
dataset = load_dataset("MarcoLents/HaLO")

pre_study

Pairwise legibility annotations from the annotation pre-study, with the same schema as the main study. Splits correspond to individual annotators, where Annotator 1 performed the procedure twice (splits annotator11 and annotator12).

Splits: annotator11 (1,839), annotator12 (1,530), annotator2 (1,606), annotator3 (1,541)

pre_study = load_dataset("MarcoLents/HaLO", "pre_study")

characteristics

Binary sample characteristics annotated per sample.

Field Description
sampleId Sample ID
text Transcribed text content
referenceSentenceId Reference sentence ID
isWrittenInPureCursive Whether the sample is written entirely in cursive
isStrokeThin Whether the stroke width is predominantly thin
containsTypo Whether at least one typographical error is present
containsCorrection Whether at least one correction is present
samplePath Path to the feature .npy file

Splits: train (1,134), test (391), validation (382)

characteristics = load_dataset("MarcoLents/HaLO", "characteristics")

absolute_pre_study

Absolute legibility ratings on a 5-point Likert scale (0 = very legible, 4 = not legible at all) from the annotation pre-study. Four annotators independently rated a random subset of 136 samples. Annotator 4 rated twice (annotator41, annotator42).

Field Description
sampleId Sample ID
referenceSentenceId Reference sentence ID
annotator41 -- annotator7 Integer Likert ratings (0--4) per annotator
samplePath Path to the feature .npy file

Splits: train (136)

absolute_pre_study = load_dataset("MarcoLents/HaLO", "absolute_pre_study")

absolute_test

Absolute legibility ratings on a 5-point Likert scale (0 = very legible, 4 = not legible at all) from the test set. Four annotators independently rated samples from the test partition. Not all annotators rated every sample; missing values indicate unrated samples. Where a rater provided multiple ratings for the same sample, the value is the rounded mean.

Note: The absolute_pre_study and absolute_test configs use different annotator numbering schemes because they originate from separate annotation campaigns described in different sections of the paper. Some of the underlying raters overlap between the two configs, but the pseudonyms are independent.

Field Description
sampleId Sample ID
referenceSentenceId Reference sentence ID
annotator8 -- annotator11 Integer Likert ratings (0--4) per annotator, NaN if not rated
samplePath Path to the feature .npy file

Splits: train (391)

absolute_test = load_dataset("MarcoLents/HaLO", "absolute_test")

aspect_ratio

Image aspect ratio for each sample.

Field Description
sampleId Sample ID
aspectRatio Width-to-height ratio of the original image
samplePath Path to the feature .npy file

Splits: train (1,134), test (391), validation (382)

aspect_ratio = load_dataset("MarcoLents/HaLO", "aspect_ratio")

Feature Files

Pixtral 12B image embeddings are stored as .npy files under features/{referenceSentenceId}/{sampleId}.npy. Each file contains a 1024-dimensional float32 vector obtained by mean-pooling the token-level outputs of the frozen Pixtral-ViT encoder.

To download the feature files:

from huggingface_hub import snapshot_download
snapshot_download(
    repo_id="MarcoLents/HaLO",
    repo_type="dataset",
    allow_patterns="features/*/*.npy",
    local_dir="data"
)

Citation

If you use this dataset, please cite:

@article{bauer2025halo,
  title={Ranking handwriting images like a human: AI-based legibility assessment by comparative ranking on the HaLO dataset},
  author={Bauer, Meike and Lents, Marco and Schmidt, Erik and Hamann, Tim and Pieger, Lukas and Salata, Susanne and Di Salvo, Francesco and Hoffmann, Tal and Barth, Jens and Ledig, Christian},
  year={2025}
}

License

This dataset is released under the MIT License.

Acknowledgements

This study was supported by the Hightech Agenda Bayern (HTA) of the Free State of Bavaria, Germany. Additional support was provided by the Bavarian State Ministry of Economic Affairs, Regional Development and Energy (StMWi) as part of the Bavarian Collaborative Research Programme (BayVFP), funding line Digitalization, project KIBEL, grant number DIK0813.

Downloads last month
976