Universe

CETA Computer Vision Dataset

byCETATask:
Object Detection
License:MIT
See all 85k imagesCETA objects sample showing personCETA objects sample showing bicycle, car, crosswalk, motorcycleCETA objects sample showing bus, car, crosswalkCETA objects sample showing car, personCETA objects sample showing personCETA objects sample showing person, truckCETA objects sample showing personCETA objects sample showing personCETA objects sample showing personCETA objects sample showing personCETA objects sample showing car, crosswalkCETA objects sample showing bus, person

About CETA Dataset

ADAS Hazards and Traffic Participants, Unified Detection Dataset

Overview

Single project that unifies COCO traffic participants with Roboflow hazards for ADAS research and embedded deployment. Images are resized to 640×640. Non-target COCO categories are removed. Labels from both sources were corrected where needed, then mapped into one label space.

Label schema, 13 classes

person, bicycle, car, bus, motorcycle, truck, cat, dog, horse, cow, speed_bump, pothole, crosswalk.

Data sources

  • COCO subset for traffic participants, classes listed above.
  • Roboflow Universe projects for static hazards, see References.

Volume

85,000 images total across COCO participants and Roboflow hazards.

Splits

Train 80%, Val 10%, Test 10%. Note, the source text listed 80%, 20%, 10%. This sums to 110%. The split here corrects that inconsistency.

Preprocessing

  • Resize to 640×640 for training and inference.
  • COCO annotations used as provided for retained classes, others dropped.
  • Roboflow annotations adopted as provided, with manual fixes for missing instances and inconsistencies to ensure one harmonized label set.

Intended use

Research on ADAS perception, hazard awareness, and multi-class detection on embedded GPUs or edge CPUs. Suitable for training detectors such as YOLO-family models and for benchmarking mixed “participants plus hazards” scenarios.

Known limitations

  • COCO images reflect diverse, general-purpose contexts, which can differ from onboard driving viewpoints.
  • Hazard datasets vary in capture conditions and geographic markings. Domain shift to your camera and region should be expected. Calibrate with targeted fine-tuning.

Licensing

  • COCO, see cocodataset.org for image licenses, COCO annotations under CC BY 4.0.
  • Roboflow Universe sources are CC BY 4.0 as listed on each project page. Verify downstream usage complies with all upstream licenses.

How to use

  1. Import into Roboflow, keep class names exactly as listed.
  2. Export in YOLO format.
  3. Train your chosen detector at 640 resolution. Start with COCO-pretrained weights, then fine-tune on this unified dataset.

Citation

If you use this project, cite COCO and the referenced Roboflow datasets, plus your chosen model paper.

References

  1. Lin, T.-Y. et al. “Microsoft COCO, Common Objects in Context,” ECCV 2014. https://cocodataset.org/
  2. Speed Bump High Light, Roboflow Universe, v1.0, CC BY 4.0. https://universe.roboflow.com/meloy/speed-bump-high-light/dataset/1
  3. TA6, Speed Bump, Roboflow Universe, CC BY 4.0. https://universe.roboflow.com/slamet-kopling-6rdvb/ta6
  4. Speed Bump Detection, Roboflow Universe, v15.0, CC BY 4.0. https://universe.roboflow.com/speed-bump-detection/speed-bump-detection-se0eh/dataset/15
  5. Speed Bump, Roboflow Universe, v3.0, CC BY 4.0. https://universe.roboflow.com/uthm-vequj/speed-bump-jvyxp/dataset/3
  6. SSD 2, Speed Bump, Roboflow Universe, v1.0, CC BY 4.0. https://universe.roboflow.com/cse400/ssd_2/dataset/1
  7. Unmarked Bumps Dataset, Roboflow Universe, v3.0, CC BY 4.0. https://universe.roboflow.com/qassim-university-oxvmm/unmarked-bumbs-gzg68/dataset/3
  8. Pothole Detection Project, Roboflow Universe, v5.0, CC BY 4.0. https://universe.roboflow.com/mgit-xi6o4/pothhole-detection-project-vk926/dataset/5
  9. Pothole Detection Dataset, Roboflow Universe, v1.0, CC BY 4.0. https://universe.roboflow.com/fatemews/pothole-detection-r1koy-rz61s/dataset/1
  10. Crosswalk Dataset, Roboflow Universe, CC BY 4.0. https://universe.roboflow.com/capstone-ergxx/crosswalk-n5psa

Use Free Car, Truck and Bus Detection API

Powered by general detection model

Code
pip install inference-sdk
# 1. Import the library
from inference_sdk import InferenceHTTPClient

# 2. Connect to your workspace
client = InferenceHTTPClient(
  api_url="https://serverless.roboflow.com",
  api_key="API_KEY"
)

# 3. Run your workflow on an image
result = client.run_workflow(
  workspace_name="<YOUR_WORKSPACE>",
  workflow_id="<YOUR_WORKFLOW_ID>",
  images={
    "image": "YOUR_IMAGE.jpg"  # Path to your image file
  },
  parameters={
    "classes": "car, truck, bus, dog, cat"
  },
  use_cache=True  # cache workflow definition for 15 minutes
)

# 4. Get your results
print(result)

Run on custom image

Drop an image here or click to upload

Detecting classes:
Or try a test image 

Cite This Project

LicenseMIT

If you use this dataset in a research paper, please cite it using the following BibTeX:

@misc{ ceta-ikg6b_dataset,
  title = { CETA Dataset },
  type = { Open Source Dataset },
  author = { CETA },
  howpublished = { \url{ https://universe.roboflow.com/ceta-4f6e8/ceta-ikg6b } },
  url = { https://universe.roboflow.com/ceta-4f6e8/ceta-ikg6b },
  journal = { Roboflow Universe },
  publisher = { Roboflow },
  year = { 2026 },
  month = { jul },
  note = { visited on 2026-07-29 },
}

Similar Projects

See More