DT.
Project Deep DiveAutoMLObject DetectionComputer VisionPythonMLOps

AutoML Framework: Automated Object Detection Training Pipeline

ZazuML-based AutoML fork for object detection — automated preprocessing, hyperparameter search, mAP evaluation, and deployment-ready weight export.

4 min readBy Drake Talley
AutoML project preview

Project Summary

AutoML framework fork based on ZazuML — streamlines object detection from labeled images through automated training, evaluation, and model export.

Technical deep dive

AutoML is a fork and extension of ZazuML — an automated machine learning framework focused on object detection workflows. It streamlines the path from labeled images to trained detectors by automating preprocessing, augmentation, model selection, and evaluation steps that otherwise require repetitive notebook boilerplate. For computer vision practitioners and hiring teams evaluating MLOps maturity, this repo demonstrates pipeline thinking: repeatable training configs, experiment tracking patterns, and deployment-oriented outputs rather than one-off Kaggle notebooks.

What AutoML automates

  • Dataset ingestion and validation for object detection formats (COCO-style, YOLO-style)
  • Automated train/validation splits with reproducible random seeds
  • Hyperparameter search across detector architectures and augmentation policies
  • Model evaluation with mAP, precision-recall curves, and confusion analysis
  • Export paths for inference-ready weights and deployment artifacts
  • Configuration-driven experiments — change YAML/JSON, not scattered script edits

Object detection pipeline

Automated detection pipeline with feedback loop on evaluation metrics.

Tech stack

  • Python 3.x with PyTorch or TensorFlow backends (per configuration)
  • OpenCV for image preprocessing and augmentation
  • Standard detection metrics (mAP, IoU thresholds)
  • YAML/JSON experiment configuration
  • Optional Docker for reproducible training environments

Getting started

git clone https://github.com/cdtalley/AutoML
cd AutoML
pip install -r requirements.txt
# Configure dataset path in config.yaml, then:
python train.py --config config.yaml

Key Features & Capabilities

  • Dataset ingestion for COCO-style and YOLO-style formats
  • Automated train/validation splits with reproducible seeds
  • Hyperparameter search across architectures and augmentations
  • mAP and precision-recall evaluation with export paths

Tech Stack & Components

PythonPyTorch/TensorFlowOpenCVYAML configDocker

Getting Started

1.Train a detector

Configure dataset path in config.yaml and run training.

git clone https://github.com/cdtalley/AutoML
pip install -r requirements.txt
python train.py --config config.yaml

Frequently asked questions

What is the AutoML repository?
A ZazuML-based AutoML framework fork focused on object detection — automating dataset prep, training, hyperparameter search, and evaluation for computer vision pipelines.
What detection formats are supported?
Common formats including COCO-style JSON annotations and YOLO-style label files. Check the repository README for the latest supported ingest paths.
How does this relate to VisionDetect?
AutoML handles the training automation layer; VisionDetect demonstrates end-to-end AI computer vision application development. Both appear in Drake Talley's GitHub portfolio under cdtalley.
Is GPU required?
Recommended for training object detectors at practical speed. CPU-only runs are possible for small datasets and smoke tests.