How telotubbies/auto_label works
This repository is an offline PPE dataset-generation and model-training pipeline. The principal workflow runs from raw images through SAM labeling, manual verification, dataset preparation, YOLO training, evaluation, and deployment export. The README documents orchestration and subsystem responsibilities; the source excerpts sampled here are partial, so links between stages follow explicit README descriptions rather than inferred implementation details.
Main entry points: batch_segment.py SAM Batch Runner · pipeline_cli.py Pipeline CLI · 03_evaluate_models.py Evaluation · 01_prepare_dataset.py Dataset Preparation
Components 22
Workflow Orchestration
- Pipeline CLI
pipeline_cli.py
SAM Auto-Labeling
- SAM Batch Runner
sam3_auto_label/src/batch_segment.py - Typed Configuration
sam3_auto_label/src/config.py - SAM Inference
sam3_auto_label/src/inference.py - SAM Model
sam3_auto_label/sam3/sam3/model_builder.py - Annotation Exports
sam3_auto_label/src/exporters.py - Experiment History
sam3_auto_label/src/tracker.py - Resume Checkpoint
sam3_auto_label/src/batch_segment.py
Dataset Preparation
- Dataset Preparation
yolo26_ppe/scripts/pipeline/01_prepare_dataset.py - Versioned Dataset
YOLO Model Lifecycle
- YOLO Training
yolo26_ppe/scripts/pipeline/02_train_models.py - Production Models
Evaluation and Outputs
- Evaluation
yolo26_ppe/scripts/pipeline/03_evaluate_models.py - Failure Analysis
yolo26_ppe/scripts/pipeline/05_generate_failure_montage.py - Blur Robustness
yolo26_ppe/scripts/pipeline/06_run_blur_robustness.py - Robustness Analysis
yolo26_ppe/scripts/pipeline/07_analyze_blur_robustness.py - ONNX Export
yolo26_ppe/scripts/pipeline/04_export_and_evaluate_onnx.py - Raw Image Prediction
yolo26_ppe/scripts/pipeline/09_predict_raw_images.py - Engineering Report
yolo26_ppe/scripts/pipeline/08_generate_report_figures.py
Other components
- Operator
- Raw Images
- Manual Verification
Connections 28
- Operator → Pipeline CLI: starts workflow
- Pipeline CLI → SAM Batch Runner: dispatches SAM
- Pipeline CLI → Dataset Preparation: dispatches preparation
- Pipeline CLI → YOLO Training: dispatches training
- Pipeline CLI → Evaluation: dispatches evaluation
- Pipeline CLI → ONNX Export: dispatches export
- Pipeline CLI → Raw Image Prediction: dispatches prediction
- Raw Images → SAM Batch Runner: supplies images
- SAM Batch Runner → Typed Configuration: reads settings
- SAM Batch Runner → SAM Inference: runs inference
- SAM Inference → SAM Model: uses model
- SAM Batch Runner → Annotation Exports: serializes annotations
- SAM Batch Runner → Experiment History: records runs
- SAM Batch Runner → Resume Checkpoint: writes resume state
- Resume Checkpoint → SAM Batch Runner: resumes batch
- Annotation Exports → Manual Verification: provides annotations
- Manual Verification → Dataset Preparation: supplies verified labels
- Dataset Preparation → Versioned Dataset: creates dataset
- Versioned Dataset → YOLO Training: trains on
- YOLO Training → Production Models: produces weights
- Production Models → Evaluation: evaluates
- Evaluation → Failure Analysis: provides results
- Production Models → ONNX Export: exports models
- ONNX Export → Engineering Report: provides export results
- Evaluation → Engineering Report: provides metrics
- Versioned Dataset → Blur Robustness: tests robustness
- Blur Robustness → Robustness Analysis: provides results
- Production Models → Raw Image Prediction: runs inference