tao-train-oneformer - Train and deploy OneFormer segmentation models
Train, evaluate, export, quantize, and run inference for TAO OneFormer models supporting panoptic, instance, and semantic segmentation.
Tags
Updated: 2026-10-06Capabilities
Typical Inputs
Typical Outputs
What this skill does
- Train OneFormer models
- Evaluate segmentation models
- Export models to ONNX
- Quantize model weights
- Run segmentation inference
- Configure AutoML training
Inputs
- COCO segmentation datasets
- Model checkpoints
- Pretrained backbones
- Action specifications
- Output paths
Outputs
- Trained model checkpoints
- Evaluation metrics
- Exported ONNX models
- Quantized models
- Segmentation inference results
Requirements
- Docker
- NVIDIA Container Toolkit
- TAO runtime environment