maticmatusek/VLM_semantics_SLO_benchmark

VLM Semantics SLO Benchmark

VLM Semantics SLO is a Slovenian multimodal benchmark for studying cultural and semiotic reasoning in vision-language models. It goes beyond object recognition by asking models to interpret visual hierarchy, spatial relations, colour and mood, composition, cultural symbols, metaphor, denotation and connotation, intertextuality, communicative intent, and relevance to Slovenia.

The released JSON contains 4,950 image-level records. Every record has ten Slovenian questions, giving 49,500 image-question task slots. Images are represented by source URLs; image files are not bundled in this release.

Dataset summary

Property	Value
Image-level records	4,950
Unique image IDs	4,950
Questions per record	10
Total question slots	49,500
Observed cultural-category labels	82
Mean categories per image	2.30
Predefined train/validation/test split	No
Primary language	Slovenian (sl)

Question types

Code	Capability	Focus
SLO	Slovenian cultural relevance	Meaning for Slovenia, Slovenian identity, or local cultural context
F1	Salience and visual hierarchy	Main motif, foreground/background, and supporting elements
F2	Relations between elements	Spatial, narrative, and symbolic relations among depicted elements
F3	Colour and mood	Palette, contrast, atmosphere, and emotional effect
I1	Composition	Balance, symmetry, visual weight, empty space, and arrangement
I2	Cultural symbolism	Cultural, religious, social, or historical meaning of symbols
I3	Metaphor	Abstract ideas communicated through visual metaphor or symbolism
A1	Denotation and connotation	What is literally depicted and what it may signify
A2	Intertextuality	References to artworks, stories, media, popular culture, or recurring motifs
A3	Communicative intent	Intended message, purpose, emotional response, and target audience

Record structure

Each JSON object represents one image.

Field	Description
image_id	Unique image identifier stored as a string
image_url	URL of the source image
image_description	Optional Slovenian image description; present in 1,020 records
image_categories	One or more Slovenian cultural-taxonomy labels
SLO, F1-F3, I1-I3, A1-A3	Slovenian questions
AnswerSLO, AnswerF1-AnswerA3	Silver-standard Slovenian reference answers
F1_suitable-A3_suitable	Question-image suitability labels; not provided for SLO
text_image_correlation	Optional assessment of agreement between the available text and image

Example, shortened for readability:

{
  "image_id": "1",
  "image_url": "https://upload.wikimedia.org/.../image.JPG",
  "image_categories": ["geografija", "naselja", "umetnost"],
  "SLO": "Kaj nam ta slika pove o Sloveniji?",
  "AnswerSLO": "...",
  "F1": "Opiši glavni motiv slike in elemente, ki ga podpirajo v ozadju.",
  "AnswerF1": "...",
  "F1_suitable": "da"
}

Data construction

Candidate images and metadata were collected from cultural collections and public web sources, followed by manual inspection, merging, deduplication, filtering, and metadata normalization. Images were assigned one or more categories from a Slovenian cultural taxonomy.

For each image, one Slovenian question was selected for each of the ten reasoning types. Gemini 2.5 Flash was used during benchmark construction to produce the reference answers and auxiliary suitability/correlation labels. These answers should therefore be treated as machine-generated silver-standard references, not as independently verified human ground truth.

The image_url field preserves image provenance at the URL level, but the release does not provide complete per-image licence metadata.

Loading the data

The JSON file can always be read with the Python standard library:

import json

with open("VLM_semantics_SLO_benchmark.json", encoding="utf-8") as file:
    records = json.load(file)

print(len(records))       # 4950
print(records[0]["SLO"])

The data use a wide format. To iterate over individual image-question pairs:

QUESTION_TYPES = ["SLO", "F1", "F2", "F3", "I1", "I2", "I3", "A1", "A2", "A3"]

for record in records:
    for question_type in QUESTION_TYPES:
        example = {
            "image_id": record["image_id"],
            "image_url": record["image_url"],
            "question_type": question_type,
            "question": record[question_type],
            "answer": record[f"Answer{question_type}"],
        }

OCM Categories

Top category counts:

Category	Count
umetnost	876
zgodovina in kulturne spremembe	525
oblačila	441
geografija	407
adolescenca, odrasla doba in starost	401
okras	362
naselja	299
objekt	290
verska prepričanja	241
orodja in naprave	233
ideje o naravi in ljudeh	224
gradnja in gradbeništvo	219
rekreacija	211
religiozne prakse	208
usnje, blago in tkanine	203
življenjski standard in navade	197
medosebni odnosi	194
delo	181
znanost in humanistika	178
vedenjski procesi in osebnosti	176
cerkvene organizacije	147
potovanje in transport	147
celinski transport	143
zgodnje in pozno otroštvo	140
imenovanje, prestiž in statusna mobilnost	139
obdelava osnovnih materialov	135
besedila	134
kmetijstvo	131
komunikacija	130
skupnost	127
človeška biologija	124
informacijski viri	122
organizacija ozemlja	121
zapisi (evidenca, register)	116
komercialna zabava	113
družina	108
vladne organizacije	98
živinoreja	96
poraba hrane	96
vladne dejavnosti	94
izobraževanje	94
organizacija poslov in industrije	92
družbeno razslojevanje	92
stroji	90
transport po vodi, zraku in vesolju	90
jezik	89
trženje	87
smrt	86
sorodstvo	82
pijača in droge	80
socializacija	79
totalna kultura	76
lastnina	76
vojaška tehnologija	76
sorodstvene skupine	76
izmenjava in prenosi	75
iskanje hrane	74
izkoriščevalske dejavnosti	73
oborožene sile	73
energija in moč	71
predelava hrane	70
oprema in vzdrževanje zgradb	68
politično vedenje	66
prekrški in sankcije	66
vojna	64
družbeni problemi	64
zakon	62
spolne vloge in vprašanja	61
zdravstvo in dobrobit	61
finance	60
bolezen	60
sodstvo	59
industrija investicijskega blaga	56
raziskovalne metode	54
demografija	53
poroka	52
orientacija	51
kemijska industrija	50
števila in mere	48
arheološke analize	47
razmnoževanje	30
seks	13

Intended uses

Suitable uses include:

• evaluation of Slovenian-capable VLMs;
• research on cultural grounding and computational semiotics;
• analysis of multimodal hallucination and visual specificity;
• supervised fine-tuning after appropriate filtering and licence review;
• comparison of literal and higher-level interpretive visual reasoning.

The dataset should not be treated as an authoritative source of cultural facts or used without expert review in high-stakes educational, heritage, legal, or social applications.

Limitations and biases

• Reference answers and metadata were generated partly by an LLM and may contain factual, visual, linguistic, or cultural errors.
• Cultural interpretations can be subjective and context-dependent.
• Web-source availability and source representation are uneven.
• Remote image links can produce link rot and make exact future reproduction difficult.
• The dataset is Slovenian-focused and does not represent every Slovenian region, community, identity, historical perspective, or contested interpretation.
• Public web images can contain identifiable people or sensitive contexts; users should review their intended use.

Licensing and image rights

The metadata and generated text are released for research use under the terms selected by the dataset publisher. The source images are not covered by a single dataset-wide licence. Copyright, attribution requirements, and permitted uses remain governed by each original source.

Before downloading, redistributing, training on, or publishing an image, verify its source page, licence, terms of use, and any applicable privacy or personality rights.