🩹 Auto-Fix Datasets
Detect and repair corrupt images, invalid labels, and duplicates in one command.
ModelCubComputer Vision, Simplified
Train and deploy CV models locally. Version datasets like code. Zero cloud costs.
Skip to content
The Workflow
bash
pip install modelcub
modelcub init my-project
modelcub dataset add --path ./data --name v1
modelcub dataset fix v1 --auto
modelcub train --dataset v1 --autoWhy ModelCub?
Stop wasting time on dataset debugging. Automatically detect and fix corrupt images, invalid labels, duplicates, and format issues.
Stop paying Roboflow $8k/month. Run everything locally with zero cloud costs.
Stop losing track of datasets. Version control with git-like semantics. Visual diffs show exactly what changed.
Python SDK
python
from modelcub import Project, Dataset, Model
# Initialize
project = Project.init("my-project")
dataset = project.import_dataset("./data", name="v1")
# Fix issues
dataset.fix(auto=True)
# Train
model = Model("yolov11n", task="detect")
run = project.train(model=model, dataset=dataset, auto=True)
# Results
print(f"mAP50: {run.evaluate().map50:.3f}")Comparison
| ModelCub | Roboflow | Ultralytics | |
|---|---|---|---|
| Cost | Free | $8k+/mo | Free |
| Runs Locally | ✅ | ❌ | ✅ |
| Auto-Fix | ✅ | ❌ | ❌ |
| Version Control | ✅ | Limited | ❌ |
| Visual Diff | ✅ | ❌ | ❌ |
Use Cases
- Medical Imaging: HIPAA-compliant, on-premise training
- Startups: Save $96k/year in cloud costs
- Research: Reproducible experiments with versioned datasets
- Privacy-Sensitive: Air-gapped environments supported
Get Started
bash
pip install modelcubRead the Installation Guide or check out the Quick Start.
MIT License • GitHub • Documentation