Description
The current implementation of Grad-CAM (Gradient-weighted Class Activation Mapping) is basic. To improve transparency and explanation of AI decisions, this issue proposes an Advanced AI Explainability Dashboard.
The system will calculate and display the percentage contribution of individual regions (Gill, Eye, and Body) to the final freshness score. The fusion formula (0.50 * Body + 0.25 * Eye + 0.25 * Gill) will be mathematically decomposed to show exactly how much weight each part contributed to the final result. In addition, the frontend will feature an interactive activation layer viewer where users can toggle between different heatmaps.
Technical Implementation Details
- Contribution Parsing: Calculate contribution percentages on the backend using the score weights and return
explainability_json.
- Visual Breakdown: Draw a custom radial ring or stacked percentage bar representing relative contribution weights in
AnalysisDashboard.tsx.
- Interactive Maps: Support tab selection to let users toggle the overlay image between body, eye, and gill activation maps.
- Fallback Mode: Generate simulated overlays for crops when in demo/no-model mode so the UI remains interactive.
Description
The current implementation of Grad-CAM (Gradient-weighted Class Activation Mapping) is basic. To improve transparency and explanation of AI decisions, this issue proposes an Advanced AI Explainability Dashboard.
The system will calculate and display the percentage contribution of individual regions (Gill, Eye, and Body) to the final freshness score. The fusion formula
(0.50 * Body + 0.25 * Eye + 0.25 * Gill)will be mathematically decomposed to show exactly how much weight each part contributed to the final result. In addition, the frontend will feature an interactive activation layer viewer where users can toggle between different heatmaps.Technical Implementation Details
explainability_json.AnalysisDashboard.tsx.