Case study
InkjetOOD
A public YOLO + conditional diffusion pipeline for industrial print-quality control evaluated on FTI_Zer0P.
Problem
Transfer the generative-classification idea into a constrained industrial setting with limited data and multiple feature types.
Role
Machine vision researcher, evaluation owner, and public benchmark analyst.
Approach
- Use YOLOv8 features as structured visual input.
- Train a conditional diffusion model with feature-aware conditioning.
- Run strict 5-fold cross-validation and compare separation-loss settings under statistical correction.
Results
- 0.8673 +/- 0.0230 AUROC baseline on the public FTI_Zer0P benchmark.
- Separation loss did not significantly improve this industrial dataset.
- The public result clarified the boundary conditions of the thesis method.
Lessons
- Industrial transfer needs more than code reuse.
- Feature-type heterogeneity changes model behavior.
- A null result can still be a strong research contribution.