[DeepKnowledge1]

🔗 Related Issue

Fixes #

📝 Description

• Add end-to-end latency measurement to the AnomaVision vs. Anomalib benchmark, separating model-only inference latency from full-pipeline latency.
• Handle Anomalib ImageItem samples without relying on PyTorch's default DataLoader collation, avoiding the reported collation error during timing.
• Include end-to-end latency in benchmark reports while preserving timing anomaly detection based on outlier fraction and coefficient of variation (CV).

🔄 Type of Change

• 🐛 Bug fix (non-breaking change which fixes an issue)
• 🚀 New feature (non-breaking change which adds functionality)
• 💥 Breaking change (fix or feature that would cause existing functionality to not work as expected)
• 📖 Documentation update
• 🏗️ Infrastructure / CI/CD update

🧪 Hardware & Matrix Testing

I have successfully built and tested this code using uv on:

• anomavision[cpu] (Standard/Edge)
• anomavision[cu121] (CUDA 12.1)
• anomavision[cu124] (CUDA 12.4)
• anomavision[cu118] (CUDA 11.8)

Host OS used for testing:

• Linux / Ubuntu
• Windows (Native or WSL2)
• macOS

✅ Developer Checklist

• My code follows the core style guidelines of this project (Ruff/Black formatting).
• I have run uv run pytest and all unit tests pass locally.
• Lockfile Guard: If I added or modified a dependency in pyproject.toml, I have run uv lock --python 3.10 and committed the updated uv.lock file.
• I have added tests that prove my fix is effective or that my feature works.
• I have updated the documentation accordingly (if applicable).

📸 Screenshots / Visual Proof (Optional)

Benchmark reports include the new end-to-end latency metric alongside model-only latency and timing stability results.