Request access to this annotated dataset
We review requests manually — usually within 1-2 business days.
This is a premium annotated dataset. By requesting access, you agree to credit UtilityMeters.ai as the data source in any research, product, or publication that uses it, and confirm you will not resell or redistribute the raw files.
Log in or Sign Up to review the conditions and access this dataset content.
Annotated-Digital-LCD-Utility-Meter-And-Serial-Numbers-Dataset-Sample
DS_electro_scor_ser_labeled_v1_2026 — Free 50-File Sample
Read the consumption value on the LCD screen and the factory serial number on the casing in a single inference pass — and bind usage data straight to the right customer account, with no manual matching.
This repo is a free preview: the first 50 annotated kits from the full dataset. Download them, drop them into your training pipeline, and check the annotation quality before you buy anything.
Dataset Summary
| Kits included (this sample) | 50 (image + YOLO .txt + PASCAL VOC .xml) |
| Full dataset | 2,290 expertly labeled kits |
| Classes | 25, context-isolated |
| Resolution | 640 px, VRAM-optimized |
| Data origin | 100% real, non-synthetic, "in-the-wild" |
Dataset Structure
DS_electro_scor_ser_labeled_v1_2026_sample/
├── Images/ # JPG photos of utility meters
├── Labels_YOLO/ # normalized relative coordinates (.txt)
├── Labels_PASCAL_VOC/ # absolute pixel coordinates (.xml)
├── classes.txt # 25-class master list
└── README_Documentation.txt
Pre-formatted for Ultralytics and standard ML pipelines — clone, point your config at it, start training.
Data Fields — 25-class context-isolated taxonomy
- On-screen metrics (IDs 0–9) — digits read directly off the LCD register.
- Serial verification — a separate character space (
s0…s9,s/,s_) for hardware serial numbers. - Target zones —
comma,scoreboard(the LCD window), andsa(the serial box), for region-first, decode-second pipelines.
Keeping on-screen digits and stamped hardware characters in separate class blocks prevents the model from confusing an LCD digit with a plastic-embossed serial character.
Why This Data Trains Better Models
- Flawless context separation. Distinct classes for on-screen metrics vs. hardware text stop the model from conflating a glowing LCD digit with a stamped serial character.
- True in-the-wild resilience. Labels explicitly cover glass and display glare, low-contrast screens in unlit basements, harsh viewing-angle distortion, faded LCD segments, and weathered or tiny serial prints.
- Manufacturer diversity. Real assets from 100+ European brands across electricity, heat, water, and gas smart meters, so models generalize instead of overfitting to one vendor.
- Training-efficient by design. Standardized 640 px resolution keeps small serial fonts legible while cutting GPU overhead.
Full Dataset & Bonus Model Weights
This sample covers 50 of 2,290 total kits. The full, production-grade dataset — plus a bonus pre-trained neural network trained on these 25 context-isolated classes (mAP50-95: 94%) — is available at:
👉 https://utilitymeters.ai/datasets-utility-meters/dataset-lcd-counters
Related Datasets from UtilityMeters.ai
UtilityMeters.ai specializes exclusively in utility-meter datasets — fully annotated sets and raw unlabeled photos for teams that prefer to label in-house:
- Mechanical dials only · LCD screens only · hybrid mechanical + digital mix
- Dials + serial numbers, for full automated asset tracking
- 20,000+ raw images, from $0.35/image, minimum order 1,000
Browse everything: https://utilitymeters.ai
Licensing Information
Datasets (images & annotations) ship with full commercial rights for training proprietary ML/AI applications. Pre-trained models are supplied as a free bonus under GNU AGPL-3.0. Redistribution or resale of raw files is strictly prohibited. Full terms on the product page.
Contact
Built for ML engineers and utility companies building autonomous billing and asset-tracking infrastructure. Questions or a custom collection request: https://utilitymeters.ai
- Downloads last month
- 101