Principle:Norrrrrrr lyn WAInjectBench Ensemble Aggregation Image
| Knowledge Sources | |
|---|---|
| Domains | Ensemble_Learning, Computer_Vision, Security |
| Last Updated | 2026-02-14 16:00 GMT |
Overview
A union-based ensemble strategy that combines image detection results from multiple detectors by merging their flagged IDs to maximize recall.
Description
The image ensemble applies the same union aggregation strategy as the text ensemble, but operates on per-folder detection results from image detectors. It reads all individual detector JSONL files from the image result directory, groups by dataset (folder) name, unions the detect_ids, and recomputes TPR/FPR. The algorithm and code structure are identical to the text ensemble.
Usage
Use this as the final aggregation step after all individual image detectors have been run. The ensemble requires that individual detector results already exist as JSONL files in the image result directory.
Theoretical Basis
Same union ensemble rule as the text variant:
The algorithm is structurally identical to the text ensemble. Both text and image ensemble modules share the same logic pattern, operating on the same JSONL result schema.