Feat:AUTO Mode for YOLO Model Selection

[arpitjod]

Describe the feature

Currently, PictoPy requires users to manually select the YOLO model size (nano, small, or medium). While this offers flexibility, it requires users to understand performance–accuracy trade-offs and system limitations.

Introduce an AUTO mode that dynamically selects the most suitable YOLO model at runtime based on contextual information such as input characteristics and system resources.

Dynamic Decision Factors

When AUTO mode is enabled, the system should evaluate multiple signals before choosing the model:

• Image resolution (small images → lighter models)
• Estimated number of faces or objects
• User priority (speed vs accuracy)
• GPU availability and free memory
• Task type (face detection or object detection)

This allows the system to adapt model choice per request instead of relying on a static configuration.

Multi-Stage Selection Strategy

AUTO mode can follow a staged decision approach:

• Pre-inference checks
  Use lightweight models for small images or limited hardware resources.

• Fast probe detection
  Run a quick detection pass using a lightweight model to estimate face/object count.

• Runtime fallback
  Automatically downgrade the model if inference fails (e.g., GPU out-of-memory).

User Benefits

• Smarter default behavior without manual tuning
• Improved performance and resource utilization
• Reduced inference failures on constrained systems
• Better overall user experience, especially for new users

Backward Compatibility

• Existing manual model selection (nano, small, medium) remains unchanged
• AUTO mode is optional and opt-in
• No breaking changes to current workflows , raise an issue for gsoc style in github

Add ScreenShots

Record

• I agree to follow this project's Code of Conduct
• I want to work on this issue

[github-actions]

👋 Thanks for opening this issue! The maintainers will review it shortly. Till then, do not open any PR. This is a strict rule we like to follow around here :)

[Anamika-hub-lang]

I’d love to work on this issue.

Proposed Approach
I plan to implement an AUTO model selection mode that dynamically chooses the optimal YOLO model at runtime based on system and input characteristics, while keeping existing manual options intact.

1. Pre-Inference Analysis

• Detect image resolution and input size
• Check GPU availability, free VRAM, and CPU fallback
• Read user preference (speed vs accuracy)
• Identify task type (face detection vs object detection)

2. Model Selection Logic

• Small images / low resources → YOLO nano
• Medium images / balanced setup → YOLO small
• Large images / accuracy-focused setup → YOLO medium

3. Fast Probe Pass (Optional)

• Run a lightweight inference pass to estimate face/object count
• Upgrade or downgrade model selection accordingly

4. Runtime Fallback

• Handle GPU OOM or inference failures gracefully
• Automatically retry with a lighter model

5. Integration & Safety

• AUTO mode will be optional and backward compatible
• No breaking changes to existing workflows
• Clear logging for model selection decisions

If this sounds aligned with the project goals, I’d be happy to take this up and start working on the implementation.
Thanks for your time and consideration 🙂

[arpitjod]

Hi @rahulharpal1603 Sir, I'd like to work on this issue.