SERVICE DETAILS
YOLOv8n ONNX KServe Inference Service
Overview
OSCAR service that deploys two services; a YOLOv8n object detection model (exported to ONNX) as a KServe InferenceService and a traditional OSCAR asynchronous service to pre-process inputs and post-process outputs through MinIO buckets. When an image is uploaded to the input bucket, the script preprocesses it, calls the KServe v2 inference endpoint, applies confidence filtering and non-maximum suppression, and writes the annotated results back to MinIO.
Deployment requirements
- Memory
- 3 GiB
- Processors
- 2 vCPU
Files & artifacts
- fdl.yml (opens in a new tab)text/yaml
Service Definition (FDL)
Show content of fdl.yml
- script.sh (opens in a new tab)text/x-shellscript
Service Execution Script
Show content of script.sh
- icon.png (opens in a new tab)image/png
Service Icon
Show content of icon.png
- input.png (opens in a new tab)image/png
Sample YOLOv8n input image — PNG image uploaded to the OSCAR input bucket for the asynchronous acceptance test.
Show content of input.png
- docker/Dockerfile (opens in a new tab)text/vnd.dockerfile
YOLOv8n ONNX modelcar Dockerfile
Show content of docker/Dockerfile
- docker/Dockerfile.script (opens in a new tab)text/vnd.dockerfile
YOLOv8n OSCAR worker Dockerfile
Show content of docker/Dockerfile.script
- docker/onnx/8/yolov8n.onnx (opens in a new tab)application/onnx
YOLOv8n ONNX model
- Detection summary outputExpected artifacttext/plain
Detection summary output — Non-empty text file named *_predictions_summary.txt generated after inference.
- Annotated detection imageExpected artifactimage/jpeg
Annotated detection image — JPEG image named *_annotated.jpg generated after inference.
Usage & acceptance tests
Asynchronous YOLOv8n ONNX KServe acceptance test
Upload the supplied image, and verify that the inference produces summary and annotated-image outputs.
Upload input.png to the configured MinIO input path and wait for the asynchronous inference to finish.
oscar-cli service put-file yolov8n-onnx-kserve {source}Wait for the remote inference to produce detection outputs.
Time required: PT20SDownload the latest annotated JPEG; inspect the summary separately as documented in the README.
oscar-cli service get-file yolov8n-onnx-kserve --download-latest-into {destination}
RO-Crate metadata
- application/json
Show content of ro-crate-metadata.json