SemProbe: AI Tool for Semantic Robustness Testing in Object Detection
SemProbe is an innovative interactive tool designed to assess the semantic robustness of object detectors in safety-sensitive areas. Users can upload images for deployment, either create masks manually or through automation, and choose factors derived from operational design domains or custom prompts to execute diffusion-based controlled inpainting. The platform accommodates batch processing, parallel seed/workflow variations, and adjustable generation settings. Following each output, model inference is conducted automatically, showcasing annotated before-and-after comparisons along with performance changes. All probing activities are recorded as structured artifacts, providing traceable evidence of robustness that aligns with safety evaluation processes. The tool was showcased in hand detection for dimension saws, focusing on criteria from insurance-related testing standards.
Key facts
- SemProbe is a tool for semantic robustness probing of object detectors.
- It uses diffusion-based controlled inpainting.
- Users can upload images and create masks manually or automatically.
- Supports batch jobs and parallel seed/workflow variations.
- Automatically runs model inference after each output.
- Displays annotated before/after comparisons with performance deltas.
- All probes are logged as structured artifacts.
- Demonstrated on hand detection for dimension saws.
Entities
—