Computer vision

These examples train computer vision models on AI Runtime. They cover image classification, hyperparameter tuning with Ray Tune, and object detection.

Example	Interface	Description
Image classification using convolutional neural network	Notebook	This notebook provides a simple example of how to train a 2-D convolution neural network on AI Runtime for image classification.
CIFAR-10 hyperparameter tuning with Ray Tune	Notebook	This notebook runs concurrent fractional-GPU trials for a PyTorch image classifier and uses the asynchronous successive halving algorithm (ASHA) to stop underperforming configurations early.
Object detection using RetinaNet	Notebook	This notebook demonstrates how to train an object detection model using RetinaNet on AI Runtime.
Object detection using YOLO11n	Notebook	This notebook demonstrates how to train a YOLO11n object detection model on the COCO128 dataset using AI Runtime, with MLflow tracking and Model Serving deployment.
ResNet-50 fine-tuning on Imagenette	CLI	Use Ray Train to fine-tune ResNet-50 with one distributed data parallel worker per A10 GPU across two nodes and log metrics to MLflow.
Image classification fine-tuning with DINOv3	CLI	Fine-tune a DINOv3 vision transformer on Oxford Flowers-102 on one A10 GPU and package the selected model for inference.
Video action classification	CLI	Train a recurrent classifier on one A10 GPU to recognize actions from video frame features produced by ResNet-18.
Batch image understanding with Qwen3-VL-8B and vLLM	CLI	Use Ray Data and vLLM to caption and classify images on an 8x H100 node with structured outputs.

Example	Interface	Description
Image classification using convolutional neural network	Notebook	This notebook provides a simple example of how to train a 2-D convolution neural network on AI Runtime for image classification.
CIFAR-10 hyperparameter tuning with Ray Tune	Notebook	This notebook runs concurrent fractional-GPU trials for a PyTorch image classifier and uses the asynchronous successive halving algorithm (ASHA) to stop underperforming configurations early.
Object detection using RetinaNet	Notebook	This notebook demonstrates how to train an object detection model using RetinaNet on AI Runtime.
Object detection using YOLO11n	Notebook	This notebook demonstrates how to train a YOLO11n object detection model on the COCO128 dataset using AI Runtime, with MLflow tracking and Model Serving deployment.
ResNet-50 fine-tuning on Imagenette	CLI	Use Ray Train to fine-tune ResNet-50 with one distributed data parallel worker per A10 GPU across two nodes and log metrics to MLflow.
Image classification fine-tuning with DINOv3	CLI	Fine-tune a DINOv3 vision transformer on Oxford Flowers-102 on one A10 GPU and package the selected model for inference.
Video action classification	CLI	Train a recurrent classifier on one A10 GPU to recognize actions from video frame features produced by ResNet-18.
Batch image understanding with Qwen3-VL-8B and vLLM	CLI	Use Ray Data and vLLM to caption and classify images on an 8x H100 node with structured outputs.