Add FastAPI inference example using OpenVINO (Windows-friendly)

[uvv-01]

What does this PR do?

Adds a minimal FastAPI-based inference example using OpenVINO, aimed at helping users quickly serve models via a REST API.

resolved #33970

Why is this needed?

• There is no simple FastAPI inference example in the repo
• Users on Windows face setup friction with existing examples
• This provides a clean, Python-only starting point

What’s included

• FastAPI inference script
• ImageNet labels
• Simple project structure under examples/fastapi

Follow-ups

• Improve documentation
• Add model download automation
• Add Linux + Docker notes

[apaniukov]

Suggested change

	core = Core()
	compiled_model = None
	input_layer = None
	output_layer = None
	compiled_model = None
	output_layer = None

Not used

[apaniukov]

ValueError should be gracefully handled here, with appropriate return code, like 415 or 422.

[Copilot]

Pull request overview

This PR adds a FastAPI-based inference example using OpenVINO to help users quickly serve models via a REST API, with a focus on Windows compatibility.

Changes:

• New FastAPI inference server with image classification endpoint
• ImageNet labels file for classification results
• Updated .gitignore to exclude Python virtual environments and downloaded models

Reviewed changes

Copilot reviewed 2 out of 4 changed files in this pull request and generated 2 comments.

File	Description
examples/fastapi/fastapi_infer.py	FastAPI server implementing image classification using OpenVINO with MobileNet-v2
examples/fastapi/imagenet_labels.txt	ImageNet class labels (first 10 entries only)
.gitignore	Adds ignores for virtual environments, model downloads, and Python cache files

Comments suppressed due to low confidence (1)

examples/fastapi/imagenet_labels.txt:1

• The ImageNet labels file only contains 10 labels, but ImageNet has 1000 classes. When the model predicts a class ID >= 10, line 88 in fastapi_infer.py (IMAGENET_LABELS[idx]) will raise an IndexError. Either include all 1000 labels or add bounds checking to handle missing labels gracefully.

tench

[Copilot]

The on_event decorator is deprecated in FastAPI. Use lifespan context manager instead for application startup/shutdown logic. Replace with @asynccontextmanager and app = FastAPI(lifespan=lifespan_context).

[Copilot]

If the model file doesn't exist at this hardcoded path, the application will crash on startup with an unclear error. Add explicit file existence checking and provide a helpful error message directing users to download the model first.

Suggested change

	model_path = "public/mobilenet-v2-pytorch/FP16/mobilenet-v2-pytorch.xml"
	model_path = os.path.join(
	base_dir,
	"public",
	"mobilenet-v2-pytorch",
	"FP16",
	"mobilenet-v2-pytorch.xml",
	)
	if not os.path.exists(model_path):
	raise RuntimeError(
	f"OpenVINO model file not found at '{model_path}'. "
	"Please download the 'mobilenet-v2-pytorch' model (FP16) from the "
	"Open Model Zoo and place it at this path before starting the app."
	)

[apaniukov]

Better use Path from pathlib

[apaniukov]

You should add a readme.md file in the fastapi folder, with:

• Short description of the example
• How to get the model, so it loads from "public/mobilenet-v2-pytorch/FP16/mobilenet-v2-pytorch.xml"
• How to create a venv and install requirements
• How to start the server
• How to query the server using CLI tool and/or Swagger and example of the output

[rkazants]

[uvv-01]

Hi all,
Thank you for the reviews and feedback on this PR. I’ve created a new PR with all requested changes implemented and cleaned up commit history to avoid the previous “10000 commits / 3000+ files changed” issue:

New PR: uvv-01/openvinonew#34338

This new PR includes:

FastAPI inference example with Windows-friendly setup

README instructions for model download, virtual environment setup, server start, and example queries

Patch file with final reviewer-requested fixes

Replaced deprecated @app.on_event("startup") with FastAPI lifespan context

Removed large offline wheel files from the repository

Only includes the relevant commits for a clean PR

Please refer to the new PR for review and merging. Thank you for your time!