Object Detection Notebooks

YOLOv5-based object detection training pipeline for OpenShift/Kubernetes with Kubeflow Pipelines.

Features

• YOLOv5 object detection training
• CPU and GPU support with automatic fallback
• Kubeflow pipeline automation
• ONNX model conversion
• OpenShift/Kubernetes deployment ready

Project Structure

object-detection-notebooks/
├── model-training/
│   ├── configuration.yaml             # Dataset configuration
│   ├── data_ingestion.py              # Data preparation
│   ├── preprocessing.py               # Data preprocessing
│   ├── model_training.py              # Main training script
│   ├── model_conversion.py            # ONNX conversion
│   ├── model_upload.py                # Model upload
│   ├── model-training-cpu.pipeline    # CPU pipeline
│   ├── model-training-gpu.pipeline    # GPU pipeline
│   └── yolov5/                        # YOLOv5 framework
├── sample-images/                     # Sample images
├── object_detection.py               # Inference script
└── requirements.txt                   # Dependencies

Quick Start

Prerequisites

• OpenShift/Kubernetes cluster with Data Science Pipelines
• S3-compatible storage

Setup

1. Place training images in /data/images/train and /data/images/val
2. Update model-training/configuration.yaml with your classes:

   nc: 1
   names: ['YourObjectClass']

3. Run the appropriate pipeline:
   • model-training-gpu.pipeline for GPU training (with CPU fallback)
   • model-training-cpu.pipeline for CPU-only training

Configuration

Training Parameters

• batch_size: Training batch size (default: 64)
• epochs: Number of epochs (default: 8)
• base_model: YOLOv5 variant (default: 'yolov5n')
• sample_count: Number of samples (default: 5000)

Pipeline Stages

1. Data Ingestion: Load and prepare training data
2. Preprocessing: Image processing and augmentation
3. Model Training: YOLOv5 training with device auto-detection
4. Model Conversion: Convert to ONNX format
5. Model Upload: Upload to S3 storage

Troubleshooting

Common Issues

• GPU Detection Error: Code automatically falls back to CPU
• Memory Issues: Reduce batch_size or use CPU pipeline
• Import Errors: Dependencies auto-install during pipeline run

Debug Device Selection

import torch
print(f"CUDA available: {torch.cuda.is_available()}")
print(f"CUDA devices: {torch.cuda.device_count()}")

Usage

Training

Run the pipeline through Data Science Pipelines UI or CLI.

Inference

from object_detection import detect_objects
results = detect_objects('image.jpg', 'model.pt')

Support

• Check pipeline logs for detailed errors
• Review troubleshooting section above
• Create repository issues for bugs