Implementation:Norrrrrrr lyn WAInjectBench load detector image

Knowledge Sources	WAInjectBench
Domains	Software_Architecture, Plugin_Systems, Computer_Vision
Last Updated	2026-02-14 16:00 GMT

Overview

Concrete tool for dynamically loading image detector modules at runtime with special LLaVA variant routing, provided by the WAInjectBench main_image module.

Description

The load_detector function in main_image.py handles two cases: (1) LLaVA variants (llava-1.5-7b-prompt, llava-1.5-7b-ft) which both load the detector_image.llava module, and (2) all other detectors which load via importlib.import_module(f"detector_image.{detector_name}"). Valid detectors include: gpt-4o-prompt, jailguard, embedding-i, llava-1.5-7b-prompt, llava-1.5-7b-ft, and ensemble.

Usage

Import and call this function when you need to dynamically select an image detector at runtime based on the --detector CLI argument.

Code Reference

Source Location

• Repository: WAInjectBench
• File: main_image.py (L7-26)

Signature

def load_detector(detector_name: str):
    """
    Dynamically load the corresponding detector module.
    - For most detectors: import detector_image.{detector_name}
    - For LLaVA series (llava-1.5-7b-prompt / llava-1.5-7b-ft):
      always import detector_image.llava
    """
    import importlib

    if detector_name in ["llava-1.5-7b-prompt", "llava-1.5-7b-ft"]:
        module = importlib.import_module("detector_image.llava")
        return module

    try:
        module = importlib.import_module(f"detector_image.{detector_name}")
        return module
    except ImportError:
        raise ValueError(
            f"Detector {detector_name} not found. "
            f"Make sure you have a file under detector_image/ named {detector_name}.py"
        )

Import

from main_image import load_detector

I/O Contract

Inputs

Name	Type	Required	Description
detector_name	str	Yes	Name of the detector (e.g., "gpt-4o-prompt", "jailguard", "embedding-i", "llava-1.5-7b-prompt", "llava-1.5-7b-ft")

Outputs

Name	Type	Description
module	module	Loaded Python module with a detect(folder_path: str) -> List[int] function (or detect(folder_path, detector_name) for LLaVA)

Usage Examples

Loading an Image Detector

# Load the CLIP embedding detector
detector = load_detector("embedding-i")
detect_ids = detector.detect("data/image/malicious/attack_scenario/")

# Load a LLaVA variant (both route to the same module)
detector = load_detector("llava-1.5-7b-ft")
# LLaVA requires detector_name to be passed through
from detector_image import llava
detect_ids = llava.detect("data/image/malicious/attack/", detector_name="llava-1.5-7b-ft")

Related Pages

Implements Principle

• Principle:Norrrrrrr_lyn_WAInjectBench_Dynamic_Detector_Loading_Image

Requires Environment

• Environment:Norrrrrrr_lyn_WAInjectBench_External_Repos_Dependencies