ConSynth-X contains synthetic augmentations derived from Construction-Site (CC BY 4.0) and SODA (research-only) datasets. Access is granted on a per-request basis after review by the OpenConstruction team. Approval is typically within 5 business days.

By requesting access you agree to: (1) use the data for non-commercial research only; (2) cite the ConSynth-X paper in any derived work; (3) for any SODA-derived rows (soda_voc, soda_ktsh subsets), separately obtain permission from the original SODA dataset team before redistribution or publication of derived artifacts.

ConSynth-X (v2-strict on `main`)

ConSynth-X is a construction computer-vision robustness benchmark: paired synthetic augmentations of real construction site images under weather, lighting, scale, and domain-shift conditions, for evaluating object detectors and vision-language models (VLMs).

The main branch now hosts the v2-strict release (DINO ≥ 0.85, 37,308 paired rows). The previous v1 release (122,156 rows) is preserved on branch v1 for reproducibility.

5 source images shown across the 12 conditions — bounding boxes drawn for the cs10k subset.

v2-strict summary

Subset	Source images	Rows (× 12 conditions)
cs10k	1,586	19,032
soda_voc	608	7,296
soda_ktsh	915	10,980
Total	3,109	37,308

Every source image is fully paired across all 12 conditions (1 original + 11 synthetic), enabling rigorous cross-condition (paired) robustness analysis.

Gating rule. A source image is retained iff all 9 GATE conditions (fog_{light,medium,heavy}, rain_light, snow_{light,heavy}, night, small) achieve DINO cosine similarity ≥ 0.85 against the source. Three conditions are autopassed (rain_heavy, rain_night, snow_night) because their physics-overlay pipelines were already quality-controlled upstream.

DINO model. facebook/dinov3-vitl16-pretrain-lvd1689m — CLS token, L2-normalized cosine.

Conditions (12)

Group	Conditions
baseline	`original`
fog	`fog_light`, `fog_medium`, `fog_heavy` — Depth Anything V2 + Koschmieder physics
rain	`rain_light`, `rain_heavy` — IP2P + physics overlay
snow	`snow_light`, `snow_heavy` — IP2P + physics overlay
night	`night` — CycleGAN-Turbo day→night
night × weather	`rain_night`, `snow_night` — Order B: orig → IP2P → day2night → physics
scale	`small` — FLUX.1-Fill-dev outpainting

Source datasets

Subset	Source	License	Notes
cs10k	Construction-Site (LouisChen15)	CC BY 4.0	Bounding boxes (`objects` field, VOC schema)
soda_voc	SODA-A (VOC view)	research-only	Bounding boxes; derived users must obtain permission from the SODA team
soda_ktsh	SODA (caption/KTSH view)	research-only	Captions; derived users must obtain permission from the SODA team

Schema (per parquet row)

image — image bytes (JPEG)
image_id, source_id, source_dataset
condition, condition_labels (multi-label)
objects (cs10k / soda_voc — VOC bbox list) or captions (soda_ktsh)
rule_*_violation (cs10k — safety rule flags)
pipeline (generator metadata), quality_scores (DINO, etc.)

Pair across conditions by source_id.

Loading

from datasets import load_dataset

# v2-strict (default = main)
ds = load_dataset("openconstruction/ConSynth-X", "cs10k", split="fog_heavy")

# All conditions for a subset
ds = load_dataset("openconstruction/ConSynth-X", "soda_voc")

# v1 release (122,156 rows, full 11-condition v1)
ds_v1 = load_dataset("openconstruction/ConSynth-X", revision="v1")

Versions

Branch	Release	Rows	Notes
`main`	v2-strict (DINO ≥ 0.85)	37,308	Paired, gated for high fidelity (current default)
`v1`	v1 (2026-04-30)	122,156	Original 11-condition release; superseded

Related (relaxed-gate) releases of v2 family:

Ben11304/ConSynth-X-v2-strict — mirror of main (kept for citation links).
v2 (DINO ≥ 0.75, 99,828 rows) and v2-unfiltered (216,000 rows) — internal; reach out for access.

License

Dataset: CC BY-NC-ND 4.0
Source datasets: see original licenses above. SODA-derived rows require permission from the SODA team before redistribution.

Citation

@dataset{consynth_x_v2_strict_2026,
  title  = {ConSynth-X v2-strict: DINO-gated paired synthetic augmentation for construction vision robustness},
  author = {Duong, Viet Huy and Xiong, Ruoxin},
  year   = {2026},
  url    = {https://huggingface.co/datasets/openconstruction/ConSynth-X}
}

Acknowledgements

OpenConstruction Open Science Initiative · Construction Management Lab, College of Architecture and Environmental Design, Kent State University. Built atop Construction-Site, SODA, Depth Anything V2, InstructPix2Pix, CycleGAN-Turbo, and FLUX.1-Fill-dev.

Downloads last month: 67

Total file size:

14.9 GB

ConSynth-X — Access Request

ConSynth-X (v2-strict on main)