HIL for Few-Shot Object Detection

Interactive Simulation for Satellite Imagery Analysis with SSOD & ZSOD

TIFF Image Annotator

Interactive Annotation Workflow

Upload a satellite image, then simulate different AI detection methods (Single-Shot, Few-Shot, Zero-Shot). Use the annotation tools to correct AI predictions or add new labels, demonstrating the crucial role of human feedback.

Annotation Tools

Simulate AI Detection

HIL Performance Analytics

These charts illustrate the conceptual impact of human intervention on model performance and efficiency over simulated HIL cycles.

Model Accuracy (mAP) Improvement

Tracks the Mean Average Precision on novel classes over successive HIL cycles, showing clear performance gains.

Training Dataset Composition

Shows how human-corrected data becomes a vital part of the dataset used for fine-tuning the model.

Uncertainty Reduction Over Time

As the model learns from human feedback, the percentage of low-confidence detections it produces steadily decreases.