Environment:Cleanlab Cleanlab Image Quality Dependencies
| Knowledge Sources | |
|---|---|
| Domains | Computer_Vision, Data_Centric_AI |
| Last Updated | 2026-02-09 19:30 GMT |
Overview
Optional dependency environment extending the Datalab install with CleanVision for automated image quality checks within cleanlab dataset audits.
Description
The image quality integration allows Datalab to detect low-quality images (blurry, dark, duplicated, etc.) alongside other data quality issues. It requires the CleanVision library (>= 0.3.6) which is an independent image quality package. This functionality is only available when using HuggingFace `datasets.Dataset` objects (not plain DataFrames or dicts).
Usage
Use this environment when running Datalab on image datasets where you want automated image quality checks (blur, darkness, duplicates). Install via the `[image]` extras group. This extends the Datalab dependency environment.
System Requirements
| Category | Requirement | Notes |
|---|---|---|
| OS | Linux, macOS, or Windows | Same as core cleanlab |
| Hardware | CPU | No GPU required for image quality checks |
| Python | >= 3.10 | Same as core cleanlab |
| Disk | Moderate | Image datasets can be large |
Dependencies
System Packages
No additional system-level packages required beyond core cleanlab.
Python Packages
- All Datalab dependencies (see Environment:Cleanlab_Cleanlab_Datalab_Dependencies)
- `cleanvision` >= 0.3.6
Credentials
No credentials required.
Quick Install
pip install 'cleanlab[image]'
Code Evidence
CleanVision import gating from `cleanlab/datalab/internal/adapter/imagelab.py:49-64`:
try:
from cleanvision import Imagelab
from datasets.arrow_dataset import Dataset
if isinstance(dataset, Dataset):
imagelab = Imagelab(hf_dataset=dataset, image_key=image_key)
else:
raise ValueError(
"For now, only huggingface datasets are supported for running "
"cleanvision checks inside cleanlab."
)
except ImportError:
raise ImportError(
"Cannot import required image packages. Please install them via: "
"`pip install cleanlab[image]` or just install cleanlab with "
"all optional dependencies via: `pip install cleanlab[all]`"
)
Optional dependency definition from `setup.py:30`:
IMAGE_REQUIRE = DATALAB_REQUIRE + ["cleanvision>=0.3.6"]
Common Errors
| Error Message | Cause | Solution |
|---|---|---|
| `Cannot import required image packages` | CleanVision or datasets not installed | `pip install 'cleanlab[image]'` |
| `only huggingface datasets are supported for running cleanvision checks` | Passed a DataFrame or dict instead of HuggingFace Dataset | Convert data to `datasets.Dataset` format first |
| `cleanvision is required for correlation visualization` | CleanVision not installed but spurious correlation visualization attempted | `pip install 'cleanlab[image]'` |
Compatibility Notes
- HuggingFace datasets only: CleanVision integration currently only supports `datasets.arrow_dataset.Dataset` objects, not pandas DataFrames or plain dicts.
- cleanvision < 0.3.6: Not supported. May cause API incompatibilities.